Home Optimization effect of video data extraction and search based on Faster-RCNN hybrid model on intelligent information systems
Article Open Access

Optimization effect of video data extraction and search based on Faster-RCNN hybrid model on intelligent information systems

  • Pei Zhang and Zhengyi Chen EMAIL logo
Published/Copyright: August 18, 2025
Download Article (PDF)
Become an author with De Gruyter Brill
Submit Manuscript Author Information
Nonlinear Engineering
From the journal Nonlinear Engineering Volume 14 Issue 1

Abstract

The swift surge of video content depicts enormous challenges for intelligent information systems in extracting and searching video data. This article explores improvements achieved by introducing a Faster-RCNN hybrid model into video data extraction and search processes. We propose a novel methodology combining Faster-RCNN with adaptive feature fusion and temporal coherence modeling to enhance object detection and tracking within video streams substantially. Using a large multi-megavideo dataset called VideoDiv-10K, which is made up of 10,000 videos from categories and others made available through public datasets like ImageNet VID and MOT16, this proposed model exhibited magnificent improvements. This model, when hybrid used with 500 h of content diversity, produced an average mean average precision of 0.891 and had an improved amount by 18.7% more than traditional Faster-RCNN. The model reduced its time for computation by 32.4% and had a massive leap in the accuracy in the search by 41.2%. Key to this performance were the temporal coherence module, which does a good job in capturing dependencies within sequential frames, and the adaptive feature fusion mechanism that dynamically integrates spatial and temporal features. Those were used to attain average F1-score in object detection of 0.937 and 0.912 object tracking on some video resolutions and frame rates. The results of the experiment clearly prove the ability of our hybrid model to process large video data, with significant gains in both accuracy and speed. We then provide better performance within applications such as surveillance, autonomous driving, and content-based video retrieval. Our model maintains a high level of scalability and does not degrade for huge datasets.

Keywords: Faster-RCNN; video data extraction; object detection; temporal coherence; intelligent information systems

1 Introduction

The amount of video content has increased exponentially in recent years, which makes it a huge challenge for intelligent information systems, especially in terms of extracting, searching, and analyzing related data. Indeed, with video content spreading in so many applications, there is a growing necessity to create efficient systems that can handle vast volumes of video data in real time and extract relevant information with the right accuracy [1]. The explosion of video data requires the development of sophisticated methodologies for dealing with intrinsic complexities inherent in video information processing, such as object appearance variability, occlusions, and dynamic video sequences [2]. Object detection and tracking is a crucial component in a wide variety of applications, ranging from surveillance and self-driving cars to content-based video retrieval and augmented reality. Intelligibility in video analysis with the ability to detect and track objects within video feeds would open this realm of use to a wide range of applications, not only making these applications more effective but also pretty accurate [3]. There are some disadvantages, however, as most traditional video analysis approaches fail to adapt well to the dynamic and unpredictable nature of video data, including sudden changes in object appearance, scene variations, motion blur, and transient occlusions [4].

In such challenges, there has been great improvement in handling such situations, especially with recent advances in deep learning, specifically convolutional neural networks (CNNs). CNNs have upgraded the efficiency of learning various hierarchical data representations and have transformed many ways computer vision is implemented, ranging from image classification to object detection and semantic segmentation [5]. Among all the state-of-the-art frameworks of object detection, the most successful one is Faster-RCNN, which has been presented by Ren et al. [6], where excellent performance has been shown in the analysis of static images due to the region proposal network (RPN) that efficiently proposed regions of interest, followed by a bounding box regression and classification. Although direct applications of Faster-RCNN on static image data deliver excellent performance, they commonly fail to fully explore the temporal information inherently embedded in video sequences [7,8]. Temporal coherence in videos can further improve the context of object detection and tracking. Traditional Faster-RCNN has no explicit mechanisms to integrate this kind of temporal information; thus, its performance might be suboptimal on video analysis tasks [9,10].

Jiang and Shi [11] explore key frame extraction technology in conjunction with the Faster R-CNN algorithm applied to the analysis of video streams from traffic. Their work introduces this model to the frontiers of information technology, data communication transmission, public clouds, and hybrid clouds. The authors explain the entire architecture of the Faster R-CNN network along with studying every possible use case for the purpose of analyzing video traffic. Xin et al. [7] propose a hybrid dilated Faster R-CNN model for object detection. The pan signifies the great performance of the model resulting from the elimination of the selective search algorithm for the original Faster R-CNN in the improvement of the performance of the model. The authors further develop the study by highlighting the role of a dilated convolution function in the optimization of the model in its potential achieved in temporal action localization in video. Zhu et al. [12] made use of a Faster R-CNN-based intelligent system for dental caries detection and localization. The authors attempt to provide evidence of the network optimization effect; in this experiment, the backbone network for feature extraction is ResNet50. The authors further discuss the potential ability of Faster R-CNN to improve the precision and speed of the detection of dental caries. Mansour et al. [13] present an intelligent video anomaly-detection and classification approach using a Faster R-CNN model combined with deep reinforcement learning. This article focuses much on the ability of the agent to find the optimal action in anomaly detection by searching through a variety of states, which underlines the efficiency of the R-CNN model.

The works of Dhevanandhini and Yamuna [14] explore the growth of optimal intelligent video surveillance systems through hybrid deep learning techniques. In this given study, deep reinforcement learning is incorporated with the focus put on the human action recognition system to improve it with the usage of Faster R-CNN. New optimizations based on two searches are also stressed by the authors. Xin et al. [7] discussed the concept of employing a hybrid dilated multilayer Faster R-CNN model as a way in object detection. In their study, they aimed to investigate how this model can be effectively utilized in video surveillance as they scrutinized the improvement in performance brought about by replacing the standard selective search algorithm with a dilated convolutional function in detail. Palle and Boda [15] have recently proposed an automatic image and video object detection system on a hybrid approach, one particular combination with U-Net segmentation and Faster R-CNN. The proposal, in this context, focuses specifically on optimization through an SF-based hybrid approach concerning the number of epochs to enhance the searching ability and efficiency of such a system. In their study, Saleem et al. [4] explore the use of a Faster R-CNN model that comes up with an improved anchor box approach for detecting weeds. The authors further discuss the optimization efforts in the Faster R-CNN model related to slow speed caused by selective search. In this study, different approaches are explored and how they affect performance. Zaman et al. [16] improved Faster R-CNN model combined with neural architectural search network research on driver emotion recognition using an improved Faster R-CNN model integrated with a neural architectural search network. The emphasis of this study lies in the three-dimensional (3D) convolutional model applied to the problem for motion encoding from video shape, where the use of Adam optimizer is considered for model optimization; and the authors emphasize the potential of the improved Faster R-CNN model for accurate driver emotion recognition.

In this work, we propose a hybrid model that fills in this gap, extending the capabilities of Faster-RCNN to the video domain, focusing on the optimization of data extraction and search processes in intelligent information systems. Hence, the hybrid would in effect suggest the combination of some amount of temporal coherence with adaptive feature fusion in facilitating improvement in video object detection and tracking. The core objectives, which we address in this article, are

  1. Designing a hybrid Faster-RCNN model as a temporal coherence adaptation model with adaptive feature fusion for better object detection and tracking in videos.

  2. To examine the performance of the proposed model in terms of accuracy, computation time, and scalability over different video datasets.

  3. To discuss the optimization impacts of the hybrid model on video data extraction and search algorithms with intelligent information systems.

  4. To analyze the robustness of the model for varying video conditions such as different resolutions, frame rates, and complexity of objects.

The article is arranged in the following way. Section 2 explains our proposed methodology, which involves the architecture our hybrid Faster-RCNN model uses and techniques for modeling temporal coherence, datasets used, evaluation metrics, and adaptive feature fusion. Baseline method comparisons, as well as state-of-the-art comparisons, are drawn up on extensive results from Section 3. We conclude this article with a summary of our findings and possible directions for future research in Section 4.

2 Model architecture and implementation

2.1 Datasets

We benchmark our hybrid Faster-RCNN model on a diverse video dataset to verify robustness across a variety of domains and scenarios. The primary corpus relied on in this article was a custom-curated collection of 10,000 videos, amounting to 500 h of content. This dataset we call VideoDiv-10K comprises an incredibly diverse mix of categories: surveillance footage, sports broadcasts, nature documentaries, as well as user-generated content. Table 1 provides a fine-grained breakdown of the VideoDiv-10K dataset.

Table 1

VideoDiv-10K dataset composition

Category Number of videos Total duration (h) Average resolution
Surveillance 3,000 150 1,280 × 720
Sports 2,500 125 1,920 × 1,080
Nature 2,000 100 3,840 × 2,160
User-generated 2,500 125 1,280 × 720

Apart from VideoDiv-10K, we also evaluate our model on two public datasets

  1. ImageNet VID-Russakovsky et al. [17]. This is a big video object detection dataset, comprising 3,862 video snippets with 30 object categories.

  2. MOT16-Mil Lab and Milan et al. [18]. It is a multiple object tracking benchmark comprising challenging video sequences with many different objects and camera motions.

2.2 Hybrid Faster-RCNN architecture

The VideoDiv-10K dataset, which includes 10,000 videos and 500 h in total, is particularly diverse, consisting of various real-world settings including surveillance camera videos, sports videos, nature shows, and user-generated videos. Since it is a wide range of real-life problems we are looking at here, they are well-represented in the datasets of each category at sliding scopes and various complexities.

Our proposed hybrid Faster-RCNN model extends the conventional Faster-RCNN architecture by Ren et al. [6] with more components to process the video better. Figure 1 illustrates the architecture of our hybrid model.

Figure 1 Architecture of the hybrid Faster-RCNN model.
Figure 1

Architecture of the hybrid Faster-RCNN model.

The primary components in our hybrid model include the following:

  1. Convolutional feature extractor: To maintain the proper balance between depth and efficiency, we use ResNet-101 [19] as the backbone for feature extraction.

  2. RPN: The object proposals produced by this network are classified and further refined in the subsequent layers.

  3. Module for temporal coherence: Features of consecutive frames are used here to capture temporal relations and enforce object coherence throughout the frames.

  4. Adaptive feature fusion: This module adaptively combines both spatial and temporal features so the model, for any given object, focuses only on the relevant information.

  5. ROI pooling and classification: These layers perform the final class classification along with object detection.

2.3 Evaluation metrics

For a comprehensive assessment of our hybrid Faster-RCNN model, we use the following metrics:

  1. Mean average precision (mAP): This is the objective function used to check accuracy for the object’s detection. It is computed at different intersection over union (IoU) thresholds.

  2. Frame per second (FPS): The metric used for evaluation of the computational effectiveness of the model in processing video frames.

  3. F1-score: It is the harmonic mean of precision and recall, giving a more balanced measure of detection accuracy.

  4. Multiple object tracking accuracy (MOTA): This gives the average tracking performance, assessed using false positives, false negatives, and identity switches.

  5. Multiple object tracking precision (MOTP): The localization precision of the tracker.

  6. Search accuracy: This refers to the accuracy for video segments retrieved in search operations based on object queries.

  7. Computational time: The total time taken for video data extraction and search.

Our hybrid Faster-RCNN model is compared with the baseline and state-of-the-art methods as below: (1) Original Faster-RCNN; (2) YOLO v4; (3) SSD; (4) Mask R-CNN; (5) FGFA; and (6) D&T. Our hybrid Faster-RCNN was implemented with PyTorch 1.8.0, and training was on a cluster of 8 NVIDIA Tesla V100 GPUs. Data augmentation techniques, including random horizontal flipping, color jittering, and multi-scale training, are adopted to improve model generalization. A fivefold cross-validation strategy is adopted to ensure good robustness for the results of the evaluation.

2.4 Temporal coherence modeling (TCM)

We introduce the TCM that operates on representations from consecutive frames to effectively model temporal coherence. The module consists of a 3D convolutional network followed by the self-attention mechanism, as shown in Figure 2.

Figure 2 Temporal coherence module.
Figure 2

Temporal coherence module.

The TCM processes features from a sliding window of K frames. K is the hyperparameter and thus can be adjusted according to the specific requirements of the application. The 3D convolution is defined by the following formula:

F t = σ ( W ∗ [ F { t − K + 1 } , … , F t ] + b ) ,

where F t denotes the feature map at time t, W is a 3D convolutional kernel, * denotes the convolution operation, and σ is the activation function, which we used as ReLU in our experiments. The self-attention mechanism is then applied to capture long-range dependencies within the temporal window:

A = softmax ( Q K T / d k ) O = A V ,

where Q, K, and V are the query, key, and value matrices, which are the outputs of the 3D convolution layers. d k is the dimension of the key vectors, and O is the output of the self-attention mechanism.

To achieve effective spatial and temporal information fusion, we propose an adaptive feature fusion module. It learns to dynamically weigh the importance of spatial and temporal features for each object proposal. It can be formulated as follows:

F fused = α ⁎ F spatial + ( 1 − α ) ⁎ F temporal ,

where F spatial and F temporal are spatial and temporal feature maps, respectively, and α is a learnable parameter to decide the fusion ratio. The value of α is calculated with the help of a very small neural network with input as a concatenation of spatial and temporal features:

α = sigmoid ( W 2 ⁎ ReLU ( W 1 ⁎ [ F spatial ; F temporal ] + b 1 ) + b 2 ) .

Here, W 1, W 2, b 1, and b 2 are learnable parameters, while [;] denotes the concatenation.

2.5 Training and optimization

We train our hybrid Faster-RCNN model using a multi-task loss function combining object classification, bounding box regression, and temporal consistency losses

L = L cls + L box + λ ⁎ L temp .

Here, L cls and L box represent the original Faster-RCNN: it is classification and regression of bounding loss; L temp is the temporal consistency loss and represents the weighting factor.

The loss of temporal consistency encourages consistent object detection over frames. It is defined as follows:

L temp = ∑ { t = 1 } { T − 1 } ∑ { i = 1 } { N } ∥ f t i − f { t + 1 } i ∥ 2 2 ,

where f t i denotes the feature vector of the i-th object at time t, T is the number of frames in the sequence, and N denotes the number of objects.

The hybrid Faster-RCNN model shows its pros and cons of consuming the computation resources as data sizes increase. Although the model remains relatively consistent with mAP and FPS metrics when scaling the dataset size, a mild trade-off in memory and processing power requirements is noted. Its temporal coherence module helps it reduce the amount of parallel computation done on individual videos, keeping high performance, thanks to its adaptive feature fusion mechanism. While the memory usage is higher in comparison with the base Faster-RCNN model, since the baseline model already hits an impressive accuracy-speed ratio, the hybrid model requires more parameters and comes out to be heavier regarding memory usage. However, the stable structure of the model allows it to handle the processing of large datasets effectively, an important consideration in applications like surveillance and video streaming, where high resource usage is predicted.

We applied the Adam optimizer of Kingma and Ba (2014), with a learning rate of 1 × 10−4 and a batch size of 16 during training. We trained our video dataset for 100 epochs. Besides early stopping based on good performance on validation, Table 2 depicts key hyperparameters used in our hybrid Faster-RCNN model.

Table 2

Hyperparameters of the hybrid Faster-RCNN model

Hyperparameter Value
Backbone network ResNet-101
RPN anchor scales [32, 64, 128, 256, 512]
RPN anchor ratios [0.5, 1, 2]
Temporal window size (K) 5
Fusion network hidden units 256
Learning rate 1 × 10−4
Batch size 16
Training epochs 100
λ (Temporal loss weight) 0.1

3 Experimental findings and analysis

3.1 Object detection performance

The hybrid Faster-RCNN model performed extremely well on the VideoDiv-10K dataset and had a great improvement regarding object detection accuracy and efficiency. When measured by mAP in different IoU thresholds, the hybrid model dominated at all IoU thresholds compared to baseline methods. For instance, with a high value of mAP = 0.891, it outperformed the original Faster-RCNN by 18.7%. This significant improvement reveals that integration of temporal coherence and adaptive feature fusion is really effective in well-capturing and representing the dynamic nature of video data. Comparative analysis also revealed the supremacy of the hybrid model compared with other state-of-the-art methods, namely YOLO v4, SSD, Mask R-CNN, FGFA, and D&T, thus establishing its robustness.

Furthermore, though the hybrid Faster-RCNN model gave much better accuracy, it did not compromise on computational efficiency. It improved processing time a lot. At FPS, it witnessed 32.4% improvement over the baseline Faster-RCNN model. With this balance of high accuracy with computational efficiency, the hybrid model positions itself as a robust tool for real-time video analysis, allowing more effective and faster object detection and tracking. Table 3 describes performance measures quite clearly, indicating how the hybrid Faster-RCNN model clearly presents impressive improvements in detection accuracy and speed processing, and corresponding data is presented in Figure 3.

Table 3

Object detection performance on VideoDiv-10K

Model mAP (IoU = 0.5) mAP (IoU = 0.75) FPS
Faster-RCNN (baseline) 0.751 0.623 15.3
YOLO v4 0.793 0.658 45.7
SSD 0.722 0.591 56.2
Mask R-CNN 0.779 0.647 8.9
FGFA 0.812 0.684 12.1
D&T 0.827 0.701 18.6
Our hybrid Faster-RCNN 0.891 0.783 22.4
Figure 3 Object detection performance.
Figure 3

Object detection performance.

The hybrid Faster-RCNN model attains the highest mAP at both IoU thresholds with a discernible gain over the baseline Faster-RCNN as well as other state-of-the-art methods. The gain in performance is, however, more pronounced at the higher IoU threshold of 0.75, implying higher localization accuracy. As it does not win in FPS, our model remains in a rather good compromise between accuracy and speed, outperforming several competitors in the two respects simultaneously. Precision–recall curves of the different models on the VideoDiv-10K dataset (Figure 4). Object detection by the Precision-Recall Radar Chart for VideoDiv-10K (Figure 4). From the radar chart, it can be observed that the hybrid Faster-RCNN model normally holds a higher value of precision at all the values of recall than the baseline methods. It reflects better detection performance, and actually, the model significantly outperformed the original Faster-RCNN, YOLO v4, SSD, Mask R-CNN, FGFA, and even D&T across the board.

Figure 4 Precision–recall radar chart for object detection on VideoDiv-10K.
Figure 4

Precision–recall radar chart for object detection on VideoDiv-10K.

3.2 Temporal coherence and tracking performance

Adding this module into the hybrid Faster-RCNN model boosts significantly its capacity to track with such a consistent passing of object identities between video frames. Our experiment on the MOT16 benchmark, which can be considered a good state-of-the-art multi-object tracking dataset, gave us high score results: it scored the highest MOTA and MOTP scores obtained thus far. This demonstrates that precision and reliability of the model are enhanced. A significant improvement comes in MOTA, as the temporal coherence module appears to be highly effective at eliminating a great deal of the mistakes that tracking systems make – most prominently identity switches, where, while tracking, the system incorrectly reassigns an identity it has accounted for from one frame to the next. Qualitative images of these improvements are depicted in Figure 5 as comparisons between baseline Faster-RCNN tracking results and our enhanced hybrid model. In this regard, the hybrid model shows superior performance, and a stable object tracks even when tracking scenarios include occlusion and rapid movement. Reducing the number of identity switches ensures that the hybrid model can also provide a smoother and coherent experience for tracking; such an ability would be very vital in downstream applications, such as automated surveillance and video analytics. Thus, such quality in robust tracking brings additional value in including temporal coherence into object detection frameworks. Table 4 shows the achieved MOTA and MOTP scores for different methods:

Figure 5 Qualitative comparison of tracking results.
Figure 5

Qualitative comparison of tracking results.

Table 4

Tracking performance on MOT16

Model MOTA MOTP
Faster-RCNN (baseline) 52.3 78.1
FGFA 56.7 79.4
D&T 59.2 80.3
Our hybrid Faster-RCNN 64.8 82.7

Figure 4 shows a qualitative comparison of object tracking results between the baseline Faster-RCNN model and the hybrid model incorporating the temporal coherence module. This is a way of showing how the hybrid model tracks a better stable trace of an object’s movement and clearly reduces the identity switches, especially when the object is under challenging occlusions and rapid motion.

3.3 Adaptive feature fusion analysis

The AFF module is the richer part of our hybrid Faster-RCNN model: it combines fine-grained spatial and temporal data to increase object detection and tracking with the dynamic selection of relevance concerning spatial and temporal features for each object proposal through a learned ratio, α. First, it catenates spatial and temporal features and then feeds the outcome to a small neural network, specialized for inferring the optimal fusion ratio. As a refund, α is outputted with a sigmoid function smoothly interpolating between spatial and temporal features using learnable parameters. This sophisticated mechanism makes the model adapt flexibly to dynamics varied with objects within the video streams.

We visualized learned fusion weights as a 3D bar chart to understand and compare this adaptive fusion better. Several interesting trends can be noticed in the analysis, for example, that moving objects maintain detection accuracy better using weights with more temporal information, whereas stationary or slow-moving objects rely greatly on spatial feature extraction-signifying greater dependence on appearance-based information. This adaptive behavior underlines the capability of the AFF module in optimizing the integration of feature elements according to characteristics of object movement and video contextual information, thereby contributing to the further improved performance of our hybrid Faster-RCNN model in tasks of extracting and searching real-world video data, as shown in Figure 6.

Figure 6 3D bar plot of adaptive feature fusion weights.
Figure 6

3D bar plot of adaptive feature fusion weights.

3.4 Performance across video resolutions and frame rates

An especially tough set of evaluations tests the robustness of the hybrid Faster-RCNN model using an extensive level of video resolutions and frame rates to ensure that it is adequate in a vast array of real-world situations. Three standard resolutions, namely 720p, 1080p, and 4K are combined with frame rates of 30 FPS, 60 FPS, and 120 FPS, respectively. The performance metric used in this full test is the F1-score, which fairly balances the precision and recall of this object detection task. Table 5 reveals that the model obtains high F1-scores at all of the resolution and frame rate testing levels. It should be noted that the differences can be seen much more obviously in the case of a higher resolution and frame rate. With this increase in the visuals, the more accurate and reliable detection of an object is made possible. For example, this finer detail and smoother motion at 4k resolution and 120 FPS is a boost for better object identification than when viewed at only 720p at 30 FPS. This leaves room for the model to be utilized in high-resolution and large frame rates in applications like surveillance and self-driving. In a nutshell, the hybrid Faster-RCNN model exhibits an excellent adaptability to retain supremacy at different video settings, thus fully utilizing and robustly handling video data extraction and searching tasks.

Table 5

F1-scores for object detection across video settings

Resolution 30 FPS 60 FPS 120 FPS
720p 0.912 0.924 0.931
1080p 0.928 0.937 0.943
4k 0.935 0.941 0.946

3.5 Video data extraction and search optimization

The proposed hybrid Faster-RCNN model performs well in areas related to video data extraction and retrieval. By integrating temporal coherence and feature fusion adaptively, the redundancies among the computations between consecutive frames are nicely minimized in the adopted model, thus reducing the computed time by 32.4% compared to standard Faster-RCNN, according to Table 6. This efficiency is critical in processing a large video set, especially for real-time applications where computation is always the bottleneck. Adaptive feature fusion helps the model to self-tune during the feature extraction process with more accuracy in object detection and tracking. This accuracy drives a phenomenal 41.2% gain in search accuracy and, therefore, makes content-based video retrieval more reliable. All the tested object categories show that the model is constantly superior to the baseline; however, the gains in detection of small and partially occluded objects are especially significant and naturally prove a challenge for the conventional methods. Thus, these improvements can guarantee higher frame accuracy analysis, thereby giving rise to overall enhancement in video data extraction processes. Therefore, the further improvements to the new search optimization capabilities will not only streamline the data-processing workflows but also improve the quality and soundness of the results of searches in ways that prove very critical for applications ranging from surveillance to multimedia content management.

Table 6

Video data extraction and search performance

Metric Baseline Faster-RCNN Our hybrid Faster-RCNN Improvement (%)
Computational time (h) 73.2 49.5 32.4
Search accuracy (%) 68.7 97.0 41.2

Figure 7 shows search accuracy for video data extraction and search by using different categories. Figure 6 illustrates the search accuracy of the video data extraction system for all object categories in comparison between the baseline Faster-RCNN model and the proposed hybrid Faster-RCNN model. It clearly shows that the hybrid model outperforms the baseline Faster-RCNN model in all the categories regarding search accuracy. However, where objects are small in category or partially occluded instances, improvements are more prominent, indicating that the capability of the proposed hybrid model is enhanced in challenging scenarios. Above is the chart showing eight categories of objects: Person, Vehicle, Animal, Furniture, Electronic, Clothing, Food, and Sports Equipment. The height of each bar represents the search accuracy for that category. Accuracy values for every category are: Person: 0.92; Vehicle: 0.88; Animal: 0.85; Furniture: 0.79; Electronic: 0.76; Clothing: 0.72; Food: 0.68; Sports Equipment: 0.6. This visualization makes easy the comparison of search accuracy over objects with different types, although the most accurate is Person at 0.92 and the lowest is Sports Equipment at 0.65.

Figure 7 Search accuracy distribution across object categories.
Figure 7

Search accuracy distribution across object categories.

3.6 Ablation study

We have carried out a comprehensive ablation study in order to establish the contribution of each constituent element of our hybrid Faster-RCNN model. The study basically involved testing many variants of configurations: the baseline Faster-RCNN, the model augmented only with the temporal coherence module, the model enhanced only with the adaptive feature fusion mechanism, and finally, the complete hybrid model that comprises both modules. For performance metrics, we used mAP at an IoU threshold of 0.5, MOTA, and Search Accuracy to measure the results. All evaluations reveal that the primary role of improvement in the final result has been taken by both the temporal coherence module and adaptive feature fusion. Including the temporal coherence module significantly improved the tracking accuracy in terms of MOTA-effectiveness in preserving object coherence between frames. In the adaptive feature fusion approach, it unveiled an improved detection precision and a high search accuracy that reflects how such features can effectively combine from multiple sources of information to refine the discriminative power of the model. While only the full hybrid used both modules, the full hybrid does indicate synergistic benefits as it outperformed all the other configurations on all metrics-complementary enhancements for maximizing the extraction of video data and optimization of the search accuracy. The results of this analysis are presented in Table 7.

Table 7

Ablation study results

Model configuration mAP (IoU = 0.5) MOTA Search accuracy (%)
Baseline Faster-RCNN 0.751 52.3 68.7
+ Temporal coherence module 0.823 59.6 84.2
+ Adaptive feature fusion 0.867 62.8 92.5
Full hybrid model 0.891 64.8 97.0

3.7 Computational efficiency analysis

The hybrid Faster-RCNN model thus balances strategic computations and performance enhancement through innovative integration with the temporal coherence module and adaptive feature fusion components. There is indeed a moderate trade-off with increased memory usage relative to the baseline Faster-RCNN; however, the improvements are well-worthy for such enhanced accuracies in object detection and tracking accuracy. For example, the hybrid model operates upon video data at significantly higher speed while delivering a notable speedup without loss in detection accuracy. This computational efficiency is quite palpable in real-time applications, wherein the optimized architecture of this hybrid model enables it to better deal with high-speed video analysis tasks. Design choices have been made such that although extra parameters do make the model look heavier, yet it is light enough for easy deployment in quick response time scenarios. As illustrated in Figure 8, a very interesting comparative analysis of computational efficiency metrics across these models demonstrates that our hybrid model outperforms the other models considerably. This can be attributed to the fact that the proposed model achieves high accuracy with a high speed as well, thereby addressing an important requirement for balanced processing in video data extraction and search applications. Thus, in short, this hybrid Faster-RCNN is very robust and can give better performance at tolerable computational overhead.

Figure 8 Computational efficiency comparison.
Figure 8

Computational efficiency comparison.

3.8 Scalability assessment

We run a series of experiments on video datasets of increasing size to estimate the scalability of our model. The results are in Figure 8, which shows that our model performs robustly at a similar level as the dataset size increases. This robustness is represented through slight variances in the mAP and FPS across scale sizes of the dataset. Specifically, while we scaled up to far larger datasets, mAP is seen to only decrease marginally, which points to model object detection and classification capabilities being impacted very minimally by data volume. Similarly, the FPS metric indicates processing speeds, and it reveals only a tiny bit of degradation for extremely large datasets, pointing that the model remained throughput-optimally working with no performance degradation. This further reiterates the point that the model holds pretty good for the really huge, real applications of humongous video surveillance systems or large-scale video streaming applications where millions of gigabytes of video data are expected to be present. The scalability test thus emphasizes how reliable and robust our approach is as it yields a lot of practical utility with respect to efficiency in processing large video collections without compromising performance or speed, as shown in Figure 9.

Figure 9 Scalability assessment.
Figure 9

Scalability assessment.

3.9 Error analysis

From our experimentation results with the hybrid faster-R-CNN model, we can point out a few of the areas that would require improvement to be of full use in video data extraction and search. Error representations are shown in detail in Figure 10. From this figure, it is elaborated that false negative was overwhelmingly dominant in crowded scenes, and there were erroneous classifications of visually similar object categories. The false negative is prevalent in crowded environments where the model cannot identify all objects, and there is a huge drop in recall. That is because of complex overlaps and occlusion of objects that are hard to extract. Moreover, it often identifies objects with similar visual features, for instance, animal species that cannot be distinguished and some vehicles, and then classifies them wrongly. These misclassifications further reduce the model’s precision and show a dire need for more robust differentiation of features. In order to achieve such objectives, future work would be focused on enhancing the ability of this model to handle more complex object situations, preferably using more advanced techniques, such as attention mechanisms or multi-scale feature extraction. Further, improvements in fine-grained classification of more discriminative features and training using further datasets can help reduce the problem of visually similar object misclassifications.

Figure 10 Error analysis breakdown.
Figure 10

Error analysis breakdown.

High recall is a crucial factor in crowded situations, but object overlaps and occlusion decrease recall significantly, making false negatives a core challenge. The Faster-RCNN model in its hybrid form fails in such situations, with objects unable to be detected or misclassified, especially if the objects belong to visually similar classes. Future work should use the ability to distinguish overlapping objects. Such as using attention mechanism to locate important objects or multi-scale feature extraction to better adapt to object size and placement in a scene. Addressing this issue with fine-grained classification methods through training on larger, more complex, and diverse subsets may reduce instances of false negatives in crowded scenes and improve overall performance.

At the same time, the article could benefit from inclusion of a more balanced discussion on the limitations of the hybrid Faster-RCNN model, with respect to false negatives, due to the crowded scenes or misclassification due to objects having similar visual features in the data processed. These challenges arise from factors such as object overlap, occlusion, and low feature discrimination, which significantly cause the recall and precision to decrease.

As such, overcoming these difficulties may lead to the integration of future methodologies that employ concepts like attention mechanisms, which help in gaining concentration/regionality toward the crucial areas in the context of densely populated environments that could potentially reduce this problem of false negatives. Integrating multi-scale feature extraction techniques may improve the model’s ability to recognize smaller and partially occluded objects, further increasing robustness in complex situations. Fine-grained classification techniques that aid with feature divergence may also be employed to help decrease errors across visually similar object categories, such as those in animal classes or specific kinds of cars, trucks, and buses. Training with more diverse and specialized datasets can further improve its generalization across broader object categories and scenarios.

4 Conclusion

This article proposed a hybrid Faster-RCNN model for data extraction and search optimization of video data in intelligent information systems. Our proposed model achieves better performance on several metrics compared to the baseline methods, Faster-RCNN model, and other state-of-the-art methods. Key findings include

  1. The hybrid Faster-RCNN model achieved 0.891 mAP, which exceeded the baseline Faster-RCNN by 18.7%, and overall performance was improved.

  2. For video data extraction and search, the model provided a 32.4% error coverage rate w.r.t. computational time compared to the baseline and reduced the search CPU time by 41.2%.

  3. The temporal coherence module made a significant improvement for the tracking performance, achieving a high MOTA at 64.8 on the MOT16 dataset.

  4. Adaptive feature fusion mechanism efficiently balances spatial and temporal features, which leads to superior detection and tracking accuracy in diverse video scenarios.

  5. Evolving with high robustness across a wide range of resolutions and frame rates, this model resulted in an average object detection F1-score of 0.937.

Overall, these results indicate that the hybrid Faster-RCNN model embodies promising prospects for intelligent information systems applications, which can be extended in the context of surveillance systems, content-based video retrieval, autonomous systems, and beyond. Future work will not only investigate more advanced temporal modeling techniques to better understand the sequential patterns of temporal interaction but will also investigate multi-modal fusion approaches that integrate various data streams into a unified framework for video analytics.

Acknowledgments

The author would like to acknowledge the anonymous reviewers for their contribution to this manuscript.

  1. Funding information: Authors state no funding involved.

  2. Author contributions: All authors have accepted responsibility for the entire content of this manuscript and approved its submission.

  3. Conflict of interest: Authors state no conflict of interest.

  4. Data availability statement: All data generated or analyzed during this study are included in this published article.

References

[1] Alkentar S, Alsahwa B, Assalem A, Karakolla D. Practical comparation of the accuracy and speed of YOLO, SSD and Faster RCNN for drone detection. Iraqi J Sci. 2021;8:1–12.10.31026/j.eng.2021.08.02Search in Google Scholar

[2] Sabir MF, Mehmood I, Alsaggaf WA, Khairullah EF, Alhuraiji S, Alghamdi AS, et al. An automated real-time face mask detection system using transfer learning with Faster-RCNN in the era of the COVID-19 pandemic. Sensors. 2021;21(16):5407.Search in Google Scholar

[3] Maity S, Chakrabarti A, Bhattacharjee D. Background modeling and foreground extraction in video data using spatio-temporal region persistence features. Comput Electr Eng. 2020;81:106536.10.1016/j.compeleceng.2019.106520Search in Google Scholar

[4] Saleem MH, Potgieter J, Arif KM. Weed detection by faster RCNN model: An enhanced anchor box approach. Agronomy. 2022;12(7):1510.10.3390/agronomy12071580Search in Google Scholar

[5] Mahum R, Irtaza AM, Masood M, Nawaz M, Nazir T. Real-time object detection and classification in surveillance videos using hybrid deep learning model. In Proceedings of the 6th Multi Disciplinary Student Research International Conference (MDSRIC), Wah, Pakistan. Vol. 30, 2021.Search in Google Scholar

[6] Ren S, He K, Girshick R, Sun J. Faster R-CNN: Towards real-time object detection with region proposal networks. IEEE Trans Pattern Anal Mach Intell. 2016;39(6):1137–49.10.1109/TPAMI.2016.2577031Search in Google Scholar PubMed

[7] Xin F, Zhang H, Pan H. Hybrid dilated multilayer faster RCNN for object detection. Vis Comput. 2024;40(1):19–29.10.1007/s00371-023-02789-ySearch in Google Scholar

[8] Adams SO, Azikwe E, Zubair MA. Artificial neural network analysis of some selected kdd cup 99 dataset for intrusion detection. Acta Inform Malays. 2022;6(2):55–61. 10.26480/aim.02.2022.55.61.Search in Google Scholar

[9] Xu JY, Wang SH. The impact of users’ perception of personalized advertising on user attitudes in short video platforms. Adv Manag Sci. 2024;13(1):35–44.Search in Google Scholar

[10] Wang H, Bai X, Tao M. Underwater object detection method based on improved faster RCNN. Appl Sci. 2023;13(15):8948.10.3390/app13042746Search in Google Scholar

[11] Jiang ZG, Shi XT. Application research of key frames extraction technology combined with optimized Faster R-CNN algorithm in traffic video analysis. Complexity. 2021;2021:5541336.10.1155/2021/6620425Search in Google Scholar

[12] Zhu Y, Xu T, Peng L, Cao Y, Zhao X, Li S, et al. Faster-RCNN based intelligent detection and localization of dental caries. Displays. 2022;74:102201.10.1016/j.displa.2022.102201Search in Google Scholar

[13] Mansour RF, Escorcia-Gutierrez J, Gamarra M, Villanueva JA, Leal N. Intelligent video anomaly detection and classification using faster RCNN with deep reinforcement learning model. Image Vis Comput. 2021;112:104225.10.1016/j.imavis.2021.104229Search in Google Scholar

[14] Dhevanandhini G, Yamuna G. An optimal intelligent video surveillance system in object detection using hybrid deep learning techniques. Multimed Tools Appl. 2024;83(15):45325–49.10.1007/s11042-023-17102-1Search in Google Scholar

[15] Palle RR, Boda R. Automated image and video object detection based on hybrid heuristic-based U-net segmentation and faster region-convolutional neural network-enabled learning. Multimed Tools Appl. 2023;82(3):3479–99.10.1007/s11042-022-13216-0Search in Google Scholar

[16] Zaman K, Sun Z, Shah SM, Shoaib M, Pei L, Hussain A. Driver emotions recognition based on improved Faster R-CNN and neural architectural search network. Symmetry. 2022;14(4):687.10.3390/sym14040687Search in Google Scholar

[17] Russakovsky O, Deng J, Su H, Krause J, Satheesh S, Ma S, et al. ImageNet large scale visual recognition challenge. Int J Comput Vis. 2015;115(3):211–52.10.1007/s11263-015-0816-ySearch in Google Scholar

[18] Milan A, Leal-Taixé L, Reid I, Roth S, Schindler K. MOT16: A benchmark for multi-object tracking. arXiv:1603.00831. 2016.Search in Google Scholar

[19] He K, Zhang X, Ren S, Sun J. Deep residual learning for image recognition. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. 2016.10.1109/CVPR.2016.90Search in Google Scholar
Received: 2025-01-03
Revised: 2025-02-20
Accepted: 2025-03-30
Published Online: 2025-08-18

© 2025 the author(s), published by De Gruyter

This work is licensed under the Creative Commons Attribution 4.0 International License.

Articles in the same Issue

  1. Research Articles
  2. Generalized (ψ,φ)-contraction to investigate Volterra integral inclusions and fractal fractional PDEs in super-metric space with numerical experiments
  3. Solitons in ultrasound imaging: Exploring applications and enhancements via the Westervelt equation
  4. Stochastic improved Simpson for solving nonlinear fractional-order systems using product integration rules
  5. Exploring dynamical features like bifurcation assessment, sensitivity visualization, and solitary wave solutions of the integrable Akbota equation
  6. Research on surface defect detection method and optimization of paper-plastic composite bag based on improved combined segmentation algorithm
  7. Impact the sulphur content in Iraqi crude oil on the mechanical properties and corrosion behaviour of carbon steel in various types of API 5L pipelines and ASTM 106 grade B
  8. Unravelling quiescent optical solitons: An exploration of the complex Ginzburg–Landau equation with nonlinear chromatic dispersion and self-phase modulation
  9. Perturbation-iteration approach for fractional-order logistic differential equations
  10. Variational formulations for the Euler and Navier–Stokes systems in fluid mechanics and related models
  11. Rotor response to unbalanced load and system performance considering variable bearing profile
  12. DeepFowl: Disease prediction from chicken excreta images using deep learning
  13. Channel flow of Ellis fluid due to cilia motion
  14. A case study of fractional-order varicella virus model to nonlinear dynamics strategy for control and prevalence
  15. Multi-point estimation weldment recognition and estimation of pose with data-driven robotics design
  16. Analysis of Hall current and nonuniform heating effects on magneto-convection between vertically aligned plates under the influence of electric and magnetic fields
  17. A comparative study on residual power series method and differential transform method through the time-fractional telegraph equation
  18. Insights from the nonlinear Schrödinger–Hirota equation with chromatic dispersion: Dynamics in fiber–optic communication
  19. Mathematical analysis of Jeffrey ferrofluid on stretching surface with the Darcy–Forchheimer model
  20. Exploring the interaction between lump, stripe and double-stripe, and periodic wave solutions of the Konopelchenko–Dubrovsky–Kaup–Kupershmidt system
  21. Computational investigation of tuberculosis and HIV/AIDS co-infection in fuzzy environment
  22. Signature verification by geometry and image processing
  23. Theoretical and numerical approach for quantifying sensitivity to system parameters of nonlinear systems
  24. Chaotic behaviors, stability, and solitary wave propagations of M-fractional LWE equation in magneto-electro-elastic circular rod
  25. Dynamic analysis and optimization of syphilis spread: Simulations, integrating treatment and public health interventions
  26. Visco-thermoelastic rectangular plate under uniform loading: A study of deflection
  27. Threshold dynamics and optimal control of an epidemiological smoking model
  28. Numerical computational model for an unsteady hybrid nanofluid flow in a porous medium past an MHD rotating sheet
  29. Regression prediction model of fabric brightness based on light and shadow reconstruction of layered images
  30. Dynamics and prevention of gemini virus infection in red chili crops studied with generalized fractional operator: Analysis and modeling
  31. Qualitative analysis on existence and stability of nonlinear fractional dynamic equations on time scales
  32. Review Article
  33. Haar wavelet collocation method for existence and numerical solutions of fourth-order integro-differential equations with bounded coefficients
  34. Special Issue: Nonlinear Analysis and Design of Communication Networks for IoT Applications - Part II
  35. Silicon-based all-optical wavelength converter for on-chip optical interconnection
  36. Research on a path-tracking control system of unmanned rollers based on an optimization algorithm and real-time feedback
  37. Analysis of the sports action recognition model based on the LSTM recurrent neural network
  38. Industrial robot trajectory error compensation based on enhanced transfer convolutional neural networks
  39. Research on IoT network performance prediction model of power grid warehouse based on nonlinear GA-BP neural network
  40. Interactive recommendation of social network communication between cities based on GNN and user preferences
  41. Application of improved P-BEM in time varying channel prediction in 5G high-speed mobile communication system
  42. Construction of a BIM smart building collaborative design model combining the Internet of Things
  43. Optimizing malicious website prediction: An advanced XGBoost-based machine learning model
  44. Economic operation analysis of the power grid combining communication network and distributed optimization algorithm
  45. Sports video temporal action detection technology based on an improved MSST algorithm
  46. Internet of things data security and privacy protection based on improved federated learning
  47. Enterprise power emission reduction technology based on the LSTM–SVM model
  48. Construction of multi-style face models based on artistic image generation algorithms
  49. Special Issue: Decision and Control in Nonlinear Systems - Part II
  50. Animation video frame prediction based on ConvGRU fine-grained synthesis flow
  51. Application of GGNN inference propagation model for martial art intensity evaluation
  52. Benefit evaluation of building energy-saving renovation projects based on BWM weighting method
  53. Deep neural network application in real-time economic dispatch and frequency control of microgrids
  54. Real-time force/position control of soft growing robots: A data-driven model predictive approach
  55. Mechanical product design and manufacturing system based on CNN and server optimization algorithm
  56. Application of finite element analysis in the formal analysis of ancient architectural plaque section
  57. Research on territorial spatial planning based on data mining and geographic information visualization
  58. Fault diagnosis of agricultural sprinkler irrigation machinery equipment based on machine vision
  59. Closure technology of large span steel truss arch bridge with temporarily fixed edge supports
  60. Intelligent accounting question-answering robot based on a large language model and knowledge graph
  61. Analysis of manufacturing and retailer blockchain decision based on resource recyclability
  62. Flexible manufacturing workshop mechanical processing and product scheduling algorithm based on MES
  63. Exploration of indoor environment perception and design model based on virtual reality technology
  64. Tennis automatic ball-picking robot based on image object detection and positioning technology
  65. A new CNN deep learning model for computer-intelligent color matching
  66. Design of AR-based general computer technology experiment demonstration platform
  67. Indoor environment monitoring method based on the fusion of audio recognition and video patrol features
  68. Health condition prediction method of the computer numerical control machine tool parts by ensembling digital twins and improved LSTM networks
  69. Establishment of a green degree evaluation model for wall materials based on lifecycle
  70. Quantitative evaluation of college music teaching pronunciation based on nonlinear feature extraction
  71. Multi-index nonlinear robust virtual synchronous generator control method for microgrid inverters
  72. Manufacturing engineering production line scheduling management technology integrating availability constraints and heuristic rules
  73. Analysis of digital intelligent financial audit system based on improved BiLSTM neural network
  74. Attention community discovery model applied to complex network information analysis
  75. A neural collaborative filtering recommendation algorithm based on attention mechanism and contrastive learning
  76. Rehabilitation training method for motor dysfunction based on video stream matching
  77. Research on façade design for cold-region buildings based on artificial neural networks and parametric modeling techniques
  78. Intelligent implementation of muscle strain identification algorithm in Mi health exercise induced waist muscle strain
  79. Optimization design of urban rainwater and flood drainage system based on SWMM
  80. Improved GA for construction progress and cost management in construction projects
  81. Evaluation and prediction of SVM parameters in engineering cost based on random forest hybrid optimization
  82. Museum intelligent warning system based on wireless data module
  83. Special Issue: Nonlinear Engineering’s significance in Materials Science
  84. Experimental research on the degradation of chemical industrial wastewater by combined hydrodynamic cavitation based on nonlinear dynamic model
  85. Study on low-cycle fatigue life of nickel-based superalloy GH4586 at various temperatures
  86. Some results of solutions to neutral stochastic functional operator-differential equations
  87. Ultrasonic cavitation did not occur in high-pressure CO2 liquid
  88. Research on the performance of a novel type of cemented filler material for coal mine opening and filling
  89. Testing of recycled fine aggregate concrete’s mechanical properties using recycled fine aggregate concrete and research on technology for highway construction
  90. A modified fuzzy TOPSIS approach for the condition assessment of existing bridges
  91. Nonlinear structural and vibration analysis of straddle monorail pantograph under random excitations
  92. Achieving high efficiency and stability in blue OLEDs: Role of wide-gap hosts and emitter interactions
  93. Construction of teaching quality evaluation model of online dance teaching course based on improved PSO-BPNN
  94. Enhanced electrical conductivity and electromagnetic shielding properties of multi-component polymer/graphite nanocomposites prepared by solid-state shear milling
  95. Optimization of thermal characteristics of buried composite phase-change energy storage walls based on nonlinear engineering methods
  96. A higher-performance big data-based movie recommendation system
  97. Nonlinear impact of minimum wage on labor employment in China
  98. Nonlinear comprehensive evaluation method based on information entropy and discrimination optimization
  99. Application of numerical calculation methods in stability analysis of pile foundation under complex foundation conditions
  100. Research on the contribution of shale gas development and utilization in Sichuan Province to carbon peak based on the PSA process
  101. Characteristics of tight oil reservoirs and their impact on seepage flow from a nonlinear engineering perspective
  102. Nonlinear deformation decomposition and mode identification of plane structures via orthogonal theory
  103. Numerical simulation of damage mechanism in rock with cracks impacted by self-excited pulsed jet based on SPH-FEM coupling method: The perspective of nonlinear engineering and materials science
  104. Cross-scale modeling and collaborative optimization of ethanol-catalyzed coupling to produce C4 olefins: Nonlinear modeling and collaborative optimization strategies
  105. Unequal width T-node stress concentration factor analysis of stiffened rectangular steel pipe concrete
  106. Special Issue: Advances in Nonlinear Dynamics and Control
  107. Development of a cognitive blood glucose–insulin control strategy design for a nonlinear diabetic patient model
  108. Big data-based optimized model of building design in the context of rural revitalization
  109. Multi-UAV assisted air-to-ground data collection for ground sensors with unknown positions
  110. Design of urban and rural elderly care public areas integrating person-environment fit theory
  111. Application of lossless signal transmission technology in piano timbre recognition
  112. Application of improved GA in optimizing rural tourism routes
  113. Architectural animation generation system based on AL-GAN algorithm
  114. Advanced sentiment analysis in online shopping: Implementing LSTM models analyzing E-commerce user sentiments
  115. Intelligent recommendation algorithm for piano tracks based on the CNN model
  116. Visualization of large-scale user association feature data based on a nonlinear dimensionality reduction method
  117. Low-carbon economic optimization of microgrid clusters based on an energy interaction operation strategy
  118. Optimization effect of video data extraction and search based on Faster-RCNN hybrid model on intelligent information systems
  119. Construction of image segmentation system combining TC and swarm intelligence algorithm
  120. Particle swarm optimization and fuzzy C-means clustering algorithm for the adhesive layer defect detection
  121. Optimization of student learning status by instructional intervention decision-making techniques incorporating reinforcement learning
  122. Fuzzy model-based stabilization control and state estimation of nonlinear systems
  123. Optimization of distribution network scheduling based on BA and photovoltaic uncertainty
  124. Tai Chi movement segmentation and recognition on the grounds of multi-sensor data fusion and the DBSCAN algorithm
https://doi.org/10.1515/nleng-2025-0133
Keywords for this article
Faster-RCNN; video data extraction; object detection; temporal coherence; intelligent information systems
Creative Commons
BY 4.0

Articles in the same Issue

  1. Research Articles
  2. Generalized (ψ,φ)-contraction to investigate Volterra integral inclusions and fractal fractional PDEs in super-metric space with numerical experiments
  3. Solitons in ultrasound imaging: Exploring applications and enhancements via the Westervelt equation
  4. Stochastic improved Simpson for solving nonlinear fractional-order systems using product integration rules
  5. Exploring dynamical features like bifurcation assessment, sensitivity visualization, and solitary wave solutions of the integrable Akbota equation
  6. Research on surface defect detection method and optimization of paper-plastic composite bag based on improved combined segmentation algorithm
  7. Impact the sulphur content in Iraqi crude oil on the mechanical properties and corrosion behaviour of carbon steel in various types of API 5L pipelines and ASTM 106 grade B
  8. Unravelling quiescent optical solitons: An exploration of the complex Ginzburg–Landau equation with nonlinear chromatic dispersion and self-phase modulation
  9. Perturbation-iteration approach for fractional-order logistic differential equations
  10. Variational formulations for the Euler and Navier–Stokes systems in fluid mechanics and related models
  11. Rotor response to unbalanced load and system performance considering variable bearing profile
  12. DeepFowl: Disease prediction from chicken excreta images using deep learning
  13. Channel flow of Ellis fluid due to cilia motion
  14. A case study of fractional-order varicella virus model to nonlinear dynamics strategy for control and prevalence
  15. Multi-point estimation weldment recognition and estimation of pose with data-driven robotics design
  16. Analysis of Hall current and nonuniform heating effects on magneto-convection between vertically aligned plates under the influence of electric and magnetic fields
  17. A comparative study on residual power series method and differential transform method through the time-fractional telegraph equation
  18. Insights from the nonlinear Schrödinger–Hirota equation with chromatic dispersion: Dynamics in fiber–optic communication
  19. Mathematical analysis of Jeffrey ferrofluid on stretching surface with the Darcy–Forchheimer model
  20. Exploring the interaction between lump, stripe and double-stripe, and periodic wave solutions of the Konopelchenko–Dubrovsky–Kaup–Kupershmidt system
  21. Computational investigation of tuberculosis and HIV/AIDS co-infection in fuzzy environment
  22. Signature verification by geometry and image processing
  23. Theoretical and numerical approach for quantifying sensitivity to system parameters of nonlinear systems
  24. Chaotic behaviors, stability, and solitary wave propagations of M-fractional LWE equation in magneto-electro-elastic circular rod
  25. Dynamic analysis and optimization of syphilis spread: Simulations, integrating treatment and public health interventions
  26. Visco-thermoelastic rectangular plate under uniform loading: A study of deflection
  27. Threshold dynamics and optimal control of an epidemiological smoking model
  28. Numerical computational model for an unsteady hybrid nanofluid flow in a porous medium past an MHD rotating sheet
  29. Regression prediction model of fabric brightness based on light and shadow reconstruction of layered images
  30. Dynamics and prevention of gemini virus infection in red chili crops studied with generalized fractional operator: Analysis and modeling
  31. Qualitative analysis on existence and stability of nonlinear fractional dynamic equations on time scales
  32. Review Article
  33. Haar wavelet collocation method for existence and numerical solutions of fourth-order integro-differential equations with bounded coefficients
  34. Special Issue: Nonlinear Analysis and Design of Communication Networks for IoT Applications - Part II
  35. Silicon-based all-optical wavelength converter for on-chip optical interconnection
  36. Research on a path-tracking control system of unmanned rollers based on an optimization algorithm and real-time feedback
  37. Analysis of the sports action recognition model based on the LSTM recurrent neural network
  38. Industrial robot trajectory error compensation based on enhanced transfer convolutional neural networks
  39. Research on IoT network performance prediction model of power grid warehouse based on nonlinear GA-BP neural network
  40. Interactive recommendation of social network communication between cities based on GNN and user preferences
  41. Application of improved P-BEM in time varying channel prediction in 5G high-speed mobile communication system
  42. Construction of a BIM smart building collaborative design model combining the Internet of Things
  43. Optimizing malicious website prediction: An advanced XGBoost-based machine learning model
  44. Economic operation analysis of the power grid combining communication network and distributed optimization algorithm
  45. Sports video temporal action detection technology based on an improved MSST algorithm
  46. Internet of things data security and privacy protection based on improved federated learning
  47. Enterprise power emission reduction technology based on the LSTM–SVM model
  48. Construction of multi-style face models based on artistic image generation algorithms
  49. Special Issue: Decision and Control in Nonlinear Systems - Part II
  50. Animation video frame prediction based on ConvGRU fine-grained synthesis flow
  51. Application of GGNN inference propagation model for martial art intensity evaluation
  52. Benefit evaluation of building energy-saving renovation projects based on BWM weighting method
  53. Deep neural network application in real-time economic dispatch and frequency control of microgrids
  54. Real-time force/position control of soft growing robots: A data-driven model predictive approach
  55. Mechanical product design and manufacturing system based on CNN and server optimization algorithm
  56. Application of finite element analysis in the formal analysis of ancient architectural plaque section
  57. Research on territorial spatial planning based on data mining and geographic information visualization
  58. Fault diagnosis of agricultural sprinkler irrigation machinery equipment based on machine vision
  59. Closure technology of large span steel truss arch bridge with temporarily fixed edge supports
  60. Intelligent accounting question-answering robot based on a large language model and knowledge graph
  61. Analysis of manufacturing and retailer blockchain decision based on resource recyclability
  62. Flexible manufacturing workshop mechanical processing and product scheduling algorithm based on MES
  63. Exploration of indoor environment perception and design model based on virtual reality technology
  64. Tennis automatic ball-picking robot based on image object detection and positioning technology
  65. A new CNN deep learning model for computer-intelligent color matching
  66. Design of AR-based general computer technology experiment demonstration platform
  67. Indoor environment monitoring method based on the fusion of audio recognition and video patrol features
  68. Health condition prediction method of the computer numerical control machine tool parts by ensembling digital twins and improved LSTM networks
  69. Establishment of a green degree evaluation model for wall materials based on lifecycle
  70. Quantitative evaluation of college music teaching pronunciation based on nonlinear feature extraction
  71. Multi-index nonlinear robust virtual synchronous generator control method for microgrid inverters
  72. Manufacturing engineering production line scheduling management technology integrating availability constraints and heuristic rules
  73. Analysis of digital intelligent financial audit system based on improved BiLSTM neural network
  74. Attention community discovery model applied to complex network information analysis
  75. A neural collaborative filtering recommendation algorithm based on attention mechanism and contrastive learning
  76. Rehabilitation training method for motor dysfunction based on video stream matching
  77. Research on façade design for cold-region buildings based on artificial neural networks and parametric modeling techniques
  78. Intelligent implementation of muscle strain identification algorithm in Mi health exercise induced waist muscle strain
  79. Optimization design of urban rainwater and flood drainage system based on SWMM
  80. Improved GA for construction progress and cost management in construction projects
  81. Evaluation and prediction of SVM parameters in engineering cost based on random forest hybrid optimization
  82. Museum intelligent warning system based on wireless data module
  83. Special Issue: Nonlinear Engineering’s significance in Materials Science
  84. Experimental research on the degradation of chemical industrial wastewater by combined hydrodynamic cavitation based on nonlinear dynamic model
  85. Study on low-cycle fatigue life of nickel-based superalloy GH4586 at various temperatures
  86. Some results of solutions to neutral stochastic functional operator-differential equations
  87. Ultrasonic cavitation did not occur in high-pressure CO2 liquid
  88. Research on the performance of a novel type of cemented filler material for coal mine opening and filling
  89. Testing of recycled fine aggregate concrete’s mechanical properties using recycled fine aggregate concrete and research on technology for highway construction
  90. A modified fuzzy TOPSIS approach for the condition assessment of existing bridges
  91. Nonlinear structural and vibration analysis of straddle monorail pantograph under random excitations
  92. Achieving high efficiency and stability in blue OLEDs: Role of wide-gap hosts and emitter interactions
  93. Construction of teaching quality evaluation model of online dance teaching course based on improved PSO-BPNN
  94. Enhanced electrical conductivity and electromagnetic shielding properties of multi-component polymer/graphite nanocomposites prepared by solid-state shear milling
  95. Optimization of thermal characteristics of buried composite phase-change energy storage walls based on nonlinear engineering methods
  96. A higher-performance big data-based movie recommendation system
  97. Nonlinear impact of minimum wage on labor employment in China
  98. Nonlinear comprehensive evaluation method based on information entropy and discrimination optimization
  99. Application of numerical calculation methods in stability analysis of pile foundation under complex foundation conditions
  100. Research on the contribution of shale gas development and utilization in Sichuan Province to carbon peak based on the PSA process
  101. Characteristics of tight oil reservoirs and their impact on seepage flow from a nonlinear engineering perspective
  102. Nonlinear deformation decomposition and mode identification of plane structures via orthogonal theory
  103. Numerical simulation of damage mechanism in rock with cracks impacted by self-excited pulsed jet based on SPH-FEM coupling method: The perspective of nonlinear engineering and materials science
  104. Cross-scale modeling and collaborative optimization of ethanol-catalyzed coupling to produce C4 olefins: Nonlinear modeling and collaborative optimization strategies
  105. Unequal width T-node stress concentration factor analysis of stiffened rectangular steel pipe concrete
  106. Special Issue: Advances in Nonlinear Dynamics and Control
  107. Development of a cognitive blood glucose–insulin control strategy design for a nonlinear diabetic patient model
  108. Big data-based optimized model of building design in the context of rural revitalization
  109. Multi-UAV assisted air-to-ground data collection for ground sensors with unknown positions
  110. Design of urban and rural elderly care public areas integrating person-environment fit theory
  111. Application of lossless signal transmission technology in piano timbre recognition
  112. Application of improved GA in optimizing rural tourism routes
  113. Architectural animation generation system based on AL-GAN algorithm
  114. Advanced sentiment analysis in online shopping: Implementing LSTM models analyzing E-commerce user sentiments
  115. Intelligent recommendation algorithm for piano tracks based on the CNN model
  116. Visualization of large-scale user association feature data based on a nonlinear dimensionality reduction method
  117. Low-carbon economic optimization of microgrid clusters based on an energy interaction operation strategy
  118. Optimization effect of video data extraction and search based on Faster-RCNN hybrid model on intelligent information systems
  119. Construction of image segmentation system combining TC and swarm intelligence algorithm
  120. Particle swarm optimization and fuzzy C-means clustering algorithm for the adhesive layer defect detection
  121. Optimization of student learning status by instructional intervention decision-making techniques incorporating reinforcement learning
  122. Fuzzy model-based stabilization control and state estimation of nonlinear systems
  123. Optimization of distribution network scheduling based on BA and photovoltaic uncertainty
  124. Tai Chi movement segmentation and recognition on the grounds of multi-sensor data fusion and the DBSCAN algorithm
Sign up now to receive a 20% welcome discount
Institutional Access
How does access work?
Have an idea on how to improve our website?
Please write us.
© 2025 De Gruyter Brill
Downloaded on 15.10.2025 from https://www.degruyterbrill.com/document/doi/10.1515/nleng-2025-0133/html
Scroll to top button