Estimation model of wild fractional vegetation cover based on RGB vegetation index and its application

Shaojun Dai; Jian Zhou; Xianping Ning; Jianxin Xu; Hua Wang

doi:10.1515/geo-2022-0661

Article Open Access

Estimation model of wild fractional vegetation cover based on RGB vegetation index and its application

Shaojun Dai , Jian Zhou , Xianping Ning , Jianxin Xu and Hua Wang

Published/Copyright: August 30, 2024

Published by

Become an author with De Gruyter Brill

Submit Manuscript Author Information

From the journal Open Geosciences Volume 16 Issue 1

Abstract

An accurate survey of field vegetation information facilitates the evaluation of ecosystems and the improvement of remote sensing models. Extracting fractional vegetation cover (FVC) information using aerial images is one of the important areas of unmanned aerial vehicles. However, for a field with diverse vegetation species and a complex surface environment, FVC estimation still has difficulty guaranteeing accuracy. A segmented FVC calculation method based on a thresholding algorithm is proposed to improve the accuracy and speed of FVC estimation. The FVC estimation models were analyzed by randomly selected sample images using four vegetation indices: excess green, excess green minus excess red index, green leaf index, and red green blue vegetation index (RGBVI). The results showed that the empirical model method performed poorly (validating R ² = 0.655 to 0.768). The isodata and triangle thresholding algorithms were introduced for vegetation segmentation, and their accuracy was analyzed. The results showed that the correlation between FVC estimation under RGBVI was the highest, and the triangle and isodata thresholding algorithms were complementary in terms of vegetation recognition accuracy, based on which a segmentation method of FVC calculation combining triangle and isodata algorithms was proposed. After testing, the accuracy of the improved FVC calculation method is higher than 90%, and the vegetation recognition accuracy is improved to more than 80%. This study is a positive guide to using digital cameras in field surveys.

Keywords: unmanned aerial vehicle; fractional vegetation cover; vegetation indices; image segmentation; vegetation recognition

1 Introduction

In recent years, the application fields of unmanned aerial vehicle (UAV) remote sensing (RS) technology have been gradually broadened as the capabilities of UAVs have been enhanced, and their costs have been reduced. UAVs can be used to investigate various environmental factors such as land erosion, growth of invasive organisms, harvest prediction of agricultural products, and population surveys of endangered species [1]. Moreover, compared to traditional manual research methods, data from UAV surveys are 43–96% more accurate than those collected manually [2]. Compared with satellites and aerial RS, drones are not affected by the environment, such as clouds, and it is more convenient to obtain information and can ensure a certain timeliness [3]. UAVs collect more accurate and credible environmental information. Therefore, UAVs equipped with camera sensors are widely utilized in agriculture [4], urban surveys [5], and geomorphological surveys [6] to achieve environmental monitoring at a relatively low cost by obtaining sufficient image resolution.

Fractional vegetation cover (FVC) is defined as the vertical projection of the vegetation canopy to the ground surface and expressed as a fraction or percentage of the reference area [7]. Surveys of FVC using UAVs usually carry multispectral sensors on the UAVs to obtain multispectral vegetation indices of the surface canopy [8]. However, multispectral cameras are expensive and require stitching and alignment of captured multispectral images. With the development of computer vision technology, vegetation recognition methods based on RGB images have been developed, and their recognition accuracy has been effectively ensured. Furukawa et al. [9] compared UAVs equipped with traditional RGB cameras and multispectral cameras, and although the recognition accuracy of multispectral cameras was significantly better than that of RGB cameras, with the improvement of the quality of the images acquired by the RGB cameras, the recognition accuracy of the RGB images was comparable to the multispectral camera.

Many studies have been conducted on monitoring or estimation of FVC [10,11,12]. The two commonly used techniques are as follows: (1) estimating vegetation ground cover percentages based on visual interpretation. The main problem with such methods is that the results are too subjective, as they are based mainly on the intuition of the observer. (2) Shooting area measurements are made at or near noon. However, in addition to being time-consuming and labor-intensive, this method is influenced by weather conditions and instrument orientation during the measurement. Field measurements are subject to various uncertainties and are not efficient, especially when estimating the usable area of a household on a large scale.

Vegetation indices refer to methods that characterize vegetation information by using the reflection and absorption characteristics of green vegetation to electromagnetic waves in linear or nonlinear combinations of sensitive wavebands [13]. There are two types of vegetation indices: visible light vegetation index and visible-near-infrared vegetation index. Suitable vegetation indices can be constructed to enhance specific information in the images according to the characteristics of the identified targets. Visible images contain only three channels, red, green, and blue, and relatively few vegetation indices can be constructed. Currently, most of the methods using vegetation indices are used for FVC estimation in agricultural engineering, and Zhang et al. [14] established FVC estimation models for different crops by vegetation indices. Guo et al. [15] calculated FVC by RGB images targeting sandy areas and achieved an accurate prediction of net biomass in sandy areas. Yan et al. [16] proposed a color mixing analysis method based on hue saturation color space to solve the influence of mixed pixels in RGB images on the accurate estimation of FVC when UAV is taken at a close range. Song et al. [17] proposed an anti-shadowing algorithm for extracting vegetation in the shadow region of vegetation projection, which solved the problem that vegetation in the shadow region is difficult to identify.

In the vegetation index thresholding method, setting a segmentation threshold is also an important factor in determining its recognition accuracy, and the most commonly used methods in segmenting images are fixed thresholding [18] and clustering [19]. Yan et al. [16] used the pure image elements of vegetation near the ground as the discriminant, converted the color space of RGB images, and then used threshold segmentation to the estimated FVC. Yi [20] used the excess green (ExG) vegetation index threshold segmentation method for FVC estimation of UAV visible images through human–computer interaction, which requires manual threshold adjustment to determine the optimal segmentation threshold. The maximum interclass variance thresholding method is one of the most commonly used image segmentation methods, which automatically selects the best threshold from the image grayscale for image segmentation. The k-means algorithm is a classical clustering algorithm that can simply and effectively segment images based on their grayscale [21]. Liu et al. [22] compared several color spaces, and the results showed that Lab color space is the best choice for extracting vegetation and used a modified model to fit the vegetation and background of component a. The results showed that the average error with supervised classification methods is less than 0.035.

Segmentation of vegetation using RGB vegetation indices is simple and effective, but previous studies identified vegetation with significant color characteristics and relatively simple backgrounds. In most studies, only one vegetation type was investigated, which limits the estimation of FVC for mixed vegetation systems. In most studies, only one vegetation type is studied, which limits the FVC estimation of mixed vegetation systems. For field surveys, first, the types of vegetation may be diverse, the brightness and contrast of the UAV-acquired images may be affected by weather and light conditions, and the colors of plants are all not necessarily uniformly distributed and easily identifiable and may be affected by plant health as well as seasonal factors. For FVC calculation and vegetation identification in complex field environments, four vegetation indices, ExG, ExGR, green leaf index (GLI), and RGBVI, are selected to establish an empirical model for FVC inversion. The performance of triangle and isodata thresholding algorithms for image segmentation and vegetation extraction in complex field environments is examined.

2 Experiment

2.1 Study area

The data for this study come from a mountainous area located in Yuanmou County, Chuxiong Prefecture, Yunnan Province, which is the planned construction site for wind farm construction. The topography and ecological environment of the site were mainly surveyed before construction. Grasslands, shrubs, and trees were mainly present in the study area. In June 2022, a complete picture of the ground environment was collected at the planned construction site in different weather and at different times of the year. The ground vegetation information was obtained at 1,920 pixels × 1,080 pixels as shown in Figure 1.

Figure 1

Information of ground canopy vegetation of the study area. (a) 25°44′37″N 101°56′42″E (2022/6/12), (b) 25°44′43″N 101°55′11″E (2022/6/15), (c) 25°44′38″N 101°55′28″E (2022/6/18), and (d) 25°44′40″N 101°50′43″E (2022/6/12).

2.2 Image acquisition

The ground-truthed FVC data used in this study were calculated from aerial photographs taken by the DJI M300RTK quadrotor UAV platform. It is a vertical takeoff and landing UAV designed and manufactured by DJI Innovation Technology Co., Ltd (Shenzhen, China) (http://www.dji.com), which is capable of performing accurate flight and hovering functions. The drones carry the Zenith H20T hybrid sensor for photography, which has 20 million camera pixels and generates a central projection containing the three spectral bands of red, green, and blue, and saves it in Joint Photographic Experts Group (JPEG) format. During the test, the UAV was flown at an altitude of 70 m above the ground, and the UAV was controlled to take photographs vertically downward directly above the sample. All digital color images were saved in the appropriate folders in 32-bit or 24-bit true color, depending on the focal length and time of capture. Since the images were acquired at different viewing angles, we chose pictures taken vertically to calculate the FVC.

2.3 Image FVC analysis

Among the acquired images, 200 images with different weather, different background information, and different vegetation coverage ratios were randomly selected to evaluate the extraction performance of different threshold segmentation algorithms for vegetation and different vegetation indices. The system was calibrated before data acquisition to ensure the accuracy of the acquired data. By planning a reasonable flight path and the shooting time interval, we provide that the UAV can fully acquire the ground information during the flight.

The color thresholder is a built-in APP of MATLAB, which is currently used as a powerful tool for image processing in many studies [23,24]. It provides four color spaces, which are generally described by three relatively independent attributes. The three independent variables interact to form a spatial coordinate called the color space [25]. Being able to extract target scenes in images by adjusting the thresholds of different components, it has been used in many studies and has shown good results. The vegetation elements in the pictures were marked in color thresholder APP, and it was tested that the extraction of vegetation under Lab color space was the best, which is consistent with the results of Liu [22]. The operation process is shown in Figure 2, where the RGB color space is first converted to Lab color space, the vegetation color information in the pictures is marked separately, and the generated color segmentation threshold information is recorded in Table 1. The threshold value can clearly divide vegetation and soil, binarize the vegetation information in the map, and obtain its vegetation cover information. All the images with vegetation extracted by this method were checked manually, and the results showed no significant deviation.

Figure 2

Process of vegetation recognition with Matlab color thresholder.

Table 1

Lab color space segmentation thresholds

Channel	Max	Mini
L	93.751	5.252
A	5.883	−31.653
B	52.467	−31.582

2.4 Analysis process

In computer vision and image processing applications, color is an important factor, and some researchers use color features to distinguish green vegetation from soil backgrounds. Usually, an appropriate algorithm is used to convert the RGB image into a grayscale image. Then, a suitable threshold is selected from the generated grayscale image for vegetation extraction. This method has two main advantages: the algorithm is simple to implement, and the calculation method is stable and efficient. Moreover, the algorithm’s stability can be ensured in the face of complex scenes. In this study, the process of comparing and analyzing the accuracy of vegetation index + threshold segmentation algorithm and manual recognition of vegetation is shown in Figure 3.

Figure 3

Flow of FVC calculation compared with vegetation identification.

2.5 RGB vegetation indices

Usually, the color of each pixel in an RGB color space image is represented using a combination of three values, which correspond to the R, G, and B components, respectively. Therefore, it is a challenging task to convert the pixel color from three values of R, G, and B to one grayscale value while maintaining the discriminative power for vegetation objects. Therefore, scholars have developed various image grayscale algorithms, which can convert RGB images into grayscale images. Vegetation indices perform the grayscale conversion by changing the weights of R, G, or B values using different formulas or equations, and four typical vegetation index algorithms are shown in Table 2.

Table 2

Vegetation index calculation formulas

Vegetation indices (VI)	Abbreviations	Formulas
Excess green [26]	ExG	2 g ‒ r ‒ b ( 1 )
		g = G R + G + B ( 2 )
		r = R R + G + B ( 3 )
		b = B R + G + B ( 4 )
Excess green minus excess red index [27]	ExGR	ExG ‒ ExR ( 5 )
GLI [28]	GLI	2 × G ‒ R ‒ B 2 × G + R + B ( 6 )
Red green blue vegetation index [29]	RGBVI	G 2 ‒ R × B G 2 + R × B ( 7 )

2.6 Empirical modeling method

The empirical model is used to establish the relationship between the vegetation indices calculated from the single band or band of RS images and the actual measured data of FVC. Then, the relationship is extended to the study area, and finally, the FVC of the whole study area is obtained. In the current study, the aforementioned four typical vegetation indices are selected for polynomial fitting, and the fitting equations are as follows:

(1) FVC = a × ( VI ) 2 + b × VI + c ,

where FVC is fractional vegetation cover, VI is vegetation index, a and b are parameter estimates of polynomial fit, and c is intercept of linear and polynomial fit.

2.7 Thresholding algorithm

2.7.1 Isodata thresholding algorithm

The algorithm was originally proposed by Ridler and Calvard [30], where an initial threshold is calculated from the average gray level, and then the image is segmented into A and B. The average gray level values T0 and T1 of A and B are calculated to obtain the average gray level values T′ of T0 and T1, and iterations are kept until the convergence condition is reached. This thresholding algorithm has been covered in many studies, and it has shown good image segmentation performance [31,32].

2.7.2 Triangle thresholding algorithm

The triangle thresholding algorithm is a method for automatically solving image segmentation [33]. Initially, the method was mainly used for chromosome studies, which use histogram data and are based on a purely geometric approach to finding the optimal threshold, which is founded on the assumption that the maximum histogram peak is on the side near the brightest and then the maximum linear distance is found by triangulation, and the segmentation threshold is based on the histogram gray level corresponding to the maximum linear distance. This technique can retain the maximum image information when the grayscale histogram peaks are weak.

2.7.3 Accuracy evaluation

To verify the performance of each thresholding algorithm, the accuracy index proposed by Coy et al. [34] was used for evaluation.

(2) Accuracy = 100 × A ∩ B A ∪ B .

A represents the image pixels labeled as vegetation canopy in the visual interpretation, and B represents the image pixels labeled as plant canopy in the threshold segmentation algorithm. Accuracy number characterizes the degree of matching between visual interpretation and threshold segmentation algorithm, accuracy is between the interval [0, 1], the higher accuracy is better, and accuracy = 1 indicates exact matching and perfect segmentation.

To analyze the correlation between the inversion results of the training data and the actual vegetation cover results, the determination coefficient (R ²) and the mean absolute error (MAE) were selected as the accuracy evaluation indexes in this study. The performance of the aforementioned vegetation cover inversion model was evaluated by the values of R ² and MAE, which are calculated as follows:

(3) R 2 = 1 − ∑ i = 1 n ( y i − y ˆ i ) 2 ∑ i = 1 n ( y i − y ˆ i ) 2 ,

(4) MAE = 1 n ∑ i = 1 n | y i − y ˆ i | ,

where y _i is the measured FVC value, y ˆ i is the predicted value of FVC, and n is the number of verified data points.

3 Results and discussion

3.1 Comparison of recognition results

Isodata and triangle thresholding methods were applied to each vegetation index. The results show that there is some noise in the images, and the noise is eliminated by median filtering using a 3 × 3 neighborhood window. The results of image segmentation based on ExG vegetation indices are shown in Figure 4, in the order of MATLAB color thresholder results (Figure 4(b)), triangle thresholding calculation results (Figure 4(c)), and isodata thresholding calculation results (Figure 4(d)). The vegetation cover information is recorded in Table 3.

Figure 4

Results of vegetation identification by different threshold algorithms based on ExG. (a) Original image, (b) references, (c) triangle, and (d) isodata.

Table 3

FVC calculation results of different methods at ExG

Segmentation methods	FVC
MATLAB color thresholder app	0.6075
Triangle	0.652
Isodata	0.302

3.2 Comparative analysis of accuracy based on different vegetation index regression model methods

In this study, polynomial fits were constructed between ground-truthing data and four commonly used vegetation indices (ExG, ExGR, GLI, and RGBVI), in which the vegetation indices and the measured vegetation cover had a good fit. As shown in Table 4, the correlation coefficients were 0.658, 0.350, 0.768, and 0.655, and the root mean square error was less than 0.01, and the mean absolute errors were 0.063, 0.869, 0.058, and 0.062, respectively. MAE = 0.058, followed by ExG (R ² = 0.658, MAE = 0.063) and RGBVI (R ² = 0.655, MAE = 0.0.062). The best model for estimating FVC is FVC = − 29.845 × GLI 2 + 12.942 × GLI − 0.539 .

Table 4

Quadratic polynomial fitting relationship for different vegetation indices

VIs	Quadratic polynomial	R ²	MAE
ExG	−10.303x ² + 6.886x − 0.3114	0.658	0.063
ExGR	5.843x ² − 0.2854x + 0.3972	0.350	0.869
GLI	−29.845x ² + 12.942x − 539	0.768	0.058
RGBVI	−13.192x ² + 9.6234x − 945	0.655	0.062

3.2.1 FVC estimation based on vegetation index and threshold algorithm

From the previous analysis, it can be seen that the GLI index can estimate FVC for complex environments, but the accuracy is insufficient due to different vegetation types. Therefore, this article examines a model for estimating FVC based on different color vegetation indices as well as a threshold segmentation algorithm, which is universally applicable regardless of crop type, image capture time, and crop spatial distribution. In this study, the data were divided into four groups based on four color vegetation indices. We chose two thresholding algorithms, isodata and triangle, to segment the vegetation index-converted images and compared the calculated FVC values with the reference FVC values. Finally, we record the obtained FVC estimation results in Figure 5, and it is obvious that the FVC calculated by ExG, GLI, and RGBVI with the triangle thresholding algorithm is basically linear with the reference FVC, but the calculated FVC value is slightly larger than the reference FVC value. In contrast, the FVC calculated by the isodata thresholding algorithm showed a nonlinear relationship with the reference FVC, and the calculated FVC values deviated significantly from the reference FVC values as the FVC increased. In addition, the two thresholding algorithms significantly deviated from the reference FVC values under the ExGR vegetation index, indicating that the ExGR vegetation index is not applicable to the FVC calculation of the current environment.

Figure 5

Comparison of the calculated FVC values with reference values at different vegetation indices. (a) ExG, (b) ExGR, (c) GLI, and (d) RGBVI.

All the earlier calculated results were fitted as quadratic polynomials and recorded in Table 5, and the correlation coefficients were higher than 0.95 for all the vegetation indices using the triangle threshold algorithm except the ExGR vegetation index and higher than 0.90 for the isodata thresholding algorithm. The correlation coefficients of RGBVI were higher than those of ExG and GLI vegetation indices.

Table 5

Quadratic polynomial fit relationship of FVC estimation of different thresholding algorithms with vegetation index

VI	Threshold	Quadratic	R ²	MAE
ExG	Triangle	y = 0.095x ² + 0.895x + 0.074	0.983	0.053
ExG	Isodata	y = −1.094x ² + 1.402x − 0.037	0.904	0.166
ExGR	Triangle	y = −0.443x ² + 1.136x − 0.006	0.912	0.078
ExGR	Isodata	y = −1.068x ² + 1.342x − 0.008	0.859	0.170
GLI	Triangle	y = 0.157x ² + 0.838x − 0.081	0.982	0.052
GLI	Isodata	y = −1.072x ² + 1.4392x − 0.045	0.915	0.149
RGBVI	Triangle	y = 0.060x ² + 0.936x + 0.075	0.985	0.042
RGBVI	Isodata	y = −1.032x ² + 1.425x − 0.043	0.933	0.142

3.3 Vegetation identification accuracy evaluation by thresholding algorithm

From the results of the aforementioned analysis, it can be seen that when the vegetation environment is more complex, triangle thresholding has an obvious advantage over several other thresholding algorithms, and the computational accuracy of the two automatic thresholding algorithms is evaluated using the accuracy number. As shown in Figure 5, the recognition accuracies of isodata and triangle thresholding algorithms under ExG, GLI, and RGBVI showed complementarity, with the triangle thresholding algorithm recognition accuracies increasing with the increase of vegetation cover. In contrast, the recognition accuracies of the isodata thresholding algorithm showed a decreasing trend with the increase in vegetation cover. Using a quadratic polynomial fit regression, the approximate accuracy R ² was greater than 85%. Under ExG, GLI, and RGBVI, the intersection points of the two polynomials were 0.355, 0.386, and 0.4150, respectively, indicating that the aforementioned vegetation coverage is the cut-off for the vegetation recognition accuracy of triangle and isodata threshold algorithms, and this cut-off provides a guide to improve the vegetation recognition accuracy (Figure 6).

Figure 6

Evaluation of vegetation identification accuracy of threshold algorithm at different vegetation indices. (a) ExG, (b) ExGR, (c) GLI, and (d) RGBVI.

3.4 Establishment of segmented threshold FVC estimation model

From the results of the aforementioned analysis, it can be seen that for the three vegetation indices of ExG, GLI, and RGBVI, the accuracy of the vegetation identified by the triangle and isodata thresholding algorithms is complementary, and it can be seen from Figure 7 that the accuracy of the vegetation area identified by the isodata thresholding algorithm is higher than 0.8 compared with the reference vegetation area when the FVC is approximately less than 0.3, while the accuracy of the vegetation area identified by the triangle thresholding algorithm is higher than 0.8 compared with the reference vegetation area when the FVC is approximately greater than 0.5. Therefore, based on the data obtained from the analysis of the RGBVI, the current work takes FVC = 0.41. As the segmentation node, the process of designing the segmented threshold FVC prediction model is shown in Figure 8. First, the input RGB image is grayed out by RGBVI, and the threshold segmentation is performed based on isodata thresholding algorithm and triangle thresholding algorithm, respectively, and the FVC quantization is performed on the segmentation result to obtain the average FVC calculation result. FVC = 0.4 is used as the segmentation node; if FVC < 0.4, isodata thresholding algorithm is selected; if FVC > 0.4, then triangle thresholding algorithm is used.

Figure 7

10% error band of FVC calculation and recognition accuracy for the threshold algorithm with RGBVI.

Figure 8

Improved segmented FVC prediction model process.

3.5 Method validation

To validate the segmented threshold FVC calculation model, the proposed segmented threshold FVC calculation model was validated using the vegetation area of the Lotus Campus of Kunming University of Technology. Figure 9(a) and (d) shows the two typical working conditions of vegetation identification, Figure 9(b) and (e) shows the identification results of the triangle threshold algorithm, and Figure 9(c) and (f) shows the identification results of isodata threshold algorithm. The comparison analysis shows that when the ground canopy has multiple vegetation, the triangle thresholding algorithm retains most of the vegetation information, while the isodata thresholding algorithm generates more noise in the recognition process. Although the isodata thresholding algorithm generates more noise in the vegetation area, it still retains the main vegetation information, and the problem can be solved by a simple binarization image processing. The improved vegetation recognition results are shown in Figure 10.

Figure 9

Vegetation identification results of different threshold algorithms. (a) The original image, (b) triangle, (c) isodata, (d) the original image, (e) triangle, (f) isodata.

Figure 10

Vegetation identification results of the improved method. (a) 25°03′43″N 102°41′44″E 2022/6/25, (b) identification result, (c) 25°03′43″N 102°41′44″E 2022/6/25, and (d) identification result.

Figure 11 shows the results calculated using the selected sample images of 50 images to examine the earlier proposed segmented threshold FVC estimation method. For the validation results, the estimated FVC is basically linear with the reference FVC, and the coefficient of determination R ² reaches 0.984. The accuracy of the segmented images obtained after matching with the reference image is higher than 0.8. This indicates that the segmented threshold FVC estimation model based on the color vegetation index proposed in this article is generally applicable regardless of vegetation type, image shooting time, and crop spatial distribution universal applicability.

Figure 11

Validation of FVC estimation and vegetation identification accuracy of the improved method.

4 Conclusions

In this article, four vegetation indices, ExG, ExGR, GLI, and RGBVI, were selected to explore the FVC calculation and vegetation recognition accuracy of the empirical model method and automatic threshold algorithm for complex vegetation environments through aerial images taken by UAV and the following conclusions were drawn.

Using the empirical model method to estimate FVC for complex environments, the GLI vegetation index (R ² = 0.768, MAE = 0.058), which performs best among ExG, ExGR, GLI, and RGBVI color vegetation indices, has a very low coefficient of determination for estimating FVC by ExGR, while the MAE is too large and is not suitable for FVC estimation.
Under the complex vegetation environment, ExG, GLI, and RGBVI with isodata and triangle threshold segmentation algorithms showed similar vegetation recognition results and recognition accuracy at both high and low vegetation cover, but ExGR indices were not effective in recognizing vegetation in the face of a complex environment.
The FVC value calculated by triangle thresholding algorithm shows a linear relationship with the reference FVC value, but in the low FVC area, the vegetation area identified by triangle thresholding algorithm has a larger error compared with the reference vegetation area, while the FVC value calculated by isodata thresholding algorithm shows a nonlinear relationship with the reference FVC value, but in the low FVC area, the vegetation recognition accuracy is superior. However, it has an advantage in the accuracy of vegetation recognition.
Considering the complementary accuracy of the two thresholding algorithms, this study selected an approximate segmentation limit of 0.4 based on the RGBVI, established a segmented vegetation cover calculation and vegetation recognition model for complex vegetation environments, and tested the method by another research site, and the test results showed that the accuracy of vegetation recognition was successfully improved to more than 80% and the error of the calculated FVC results less than 10% compared with the reference FVC value.

Acknowledgments

This work is partially supported by the China Energy Construction Group Yunnan Thermal Power Construction Co. (HDWZ-HT-YQM-2022-02-01).

Author contributions: Shaojun Dai: Writing – review & editing, Data acquisition, Methodology, Conceptualization; Jian Zhou: Writing – review & editing, Software; Xianping Ning: Writing – review & editing, Validation, Formal analysis, Data curation; Jianxin Xu: Supervision, Funding acquisition, Visualization, Validation, Data curation; Hua Wang: Visualization, Supervision, Validation, Data curation.
Conflict of interest: There are no conflicts to declare.
Ethical approval: There are no researches conducted on animals or humans.

References

[1] Muchiri GN, Kimathi S. A review of applications and potential applications of UAV. Proceedings of the Sustainable Research and Innovation Conference; 2022. p. 280–3.Search in Google Scholar

[2] Tmušić G, Manfreda S, Aasen H, James MR, Gonçalves G, Ben-Dor E, et al. Current practices in UAS-based environmental monitoring. Remote Sens. 2020;12(6):1001.10.3390/rs12061001Search in Google Scholar

[3] Motohka T, Nasahara KN, Oguma H, Tsuchida S. Applicability of green-red vegetation index for remote sensing of vegetation phenology. Remote Sens. 2010;2(10):2369–87.10.3390/rs2102369Search in Google Scholar

[4] Tsouros DC, Bibi S, Sarigiannidis PG. A review on UAV-based applications for precision agriculture. Information. 2019. 10(11):349.10.3390/info10110349Search in Google Scholar

[5] Thakur N, Nagrath P, Jain R, Saini D, Sharma N, Hemanth DJ. Artificial intelligence techniques in smart cities surveillance using UAVs: A survey. Mach Intell Data Anal Sustain Future Smart Cities. 2021;971:329–53.10.1007/978-3-030-72065-0_18Search in Google Scholar

[6] Xiang J, Chen J, Sofia G, Tian Y, Tarolli P. Open-pit mine geomorphic changes analysis using multi-temporal UAV survey. Environ Earth Sci. 2018;77(6):220.10.1007/s12665-018-7383-9Search in Google Scholar

[7] Purevdorj T, Tateishi R, Ishiyama T, Honda Y. Relationships between percent vegetation cover and vegetation indices. Int J Remote Sens. 1998;19(18):3519–35.10.1080/014311698213795Search in Google Scholar

[8] Song B, Park K. Detection of aquatic plants using multispectral UAV imagery and vegetation index. Remote Sens. 2020;12(3):387.10.3390/rs12030387Search in Google Scholar

[9] Furukawa F, Laneng LA, Ando H, Yoshimura N, Kaneko M, Morimoto J. Comparison of RGB and multispectral unmanned aerial vehicle for monitoring vegetation coverage changes on a landslide area. Drones. 2021;5(3):97.10.3390/drones5030097Search in Google Scholar

[10] Wang B, Jia K, Liang S, Xie X, Wei X, Zhao X, et al. Assessment of Sentinel-2 MSI spectral band reflectances for estimating fractional vegetation cover. Remote Sens. 2018;10(12):1927.10.3390/rs10121927Search in Google Scholar

[11] Maurya AK, Bhargava N, Singh D. Efficient selection of SAR features using ML based algorithms for accurate FVC estimation. Adv Space Res. 2022;70(7):1795–809.10.1016/j.asr.2022.06.039Search in Google Scholar

[12] Wang ZX, Li JL, Cao JX, Liu YS, Li D, Zhang XY, et al. Improving estimates of grassland fractional vegetation cover based on a pixel dichotomy model: A case study in Inner Mongolia, China. Remote Sens. 2014;6(6):4705–22.10.3390/rs6064705Search in Google Scholar

[13] Ni G. Vegetation index and its advances. J Arid Meteorol. 2003;21(4):71.Search in Google Scholar

[14] Zhang D, Mansaray LR, Jin H, Sun H, Kuang Z, Huang J. A universal estimation model of fractional vegetation cover for different crops based on time series digital photographs. Comput Electron Agric. 2018;151:93–103.10.1016/j.compag.2018.05.030Search in Google Scholar

[15] Guo Z, Wang T, Liu S, Kang W, Chen X, Feng K, et al. Biomass and vegetation coverage survey in the Mu Us sandy land-based on unmanned aerial vehicle RGB images. Int J Appl Earth Obs Geoinf. 2021;94:102239.10.1016/j.jag.2020.102239Search in Google Scholar

[16] Yan G, Li L, Coy A, Mu X, Chen S, Xie D, et al. Improving the estimation of fractional vegetation cover from UAV RGB imagery by colour unmixing. ISPRS J Photogramm Remote Sens. 2019;158:23–34.10.1016/j.isprsjprs.2019.09.017Search in Google Scholar

[17] Song W, Mu X, Yan G, Huang S. Extracting the green fractional vegetation cover from digital images using a shadow-resistant algorithm (SHAR-LABFVC). Remote Sens. 2015;7(8):10425–43.10.3390/rs70810425Search in Google Scholar

[18] Aho K, Roberts DW, Weaver T. Using geometric and non‐geometric internal evaluators to compare eight vegetation classification methods. J Veg Sci. 2008;19(4):549–62.10.3170/2008-8-18406Search in Google Scholar

[19] Kim DW, Yun H, Jeong SJ, Kwon YS, Kim SG, Lee W, et al. 2018. Modeling and testing of growth status for Chinese cabbage and white radish with UAV-based RGB imagery. Remote Sens. 2018;10(4):563.10.3390/rs10040563Search in Google Scholar

[20] Yi SH. FragMAP: a tool for long term and cooperative monitoring and analysis of small scale habitat fragmentation using an unmanned aerial vehicle. Int J Remote Sens. 2017;38(8–10):2686–97.10.1080/01431161.2016.1253898Search in Google Scholar

[21] Shukla S, Naganna S. A review on K-means data clustering approach. Int J Inf Comput Technol. 2014;4(17):1847–60.Search in Google Scholar

[22] Liu Y, Mu X, Wang H, Yan G. A novel method for extracting green fractional vegetation cover from digital images. J Veg Sci. 2012;23(3):406–18.10.1111/j.1654-1103.2011.01373.xSearch in Google Scholar

[23] Sultana U, Hoque M, Akter P, Islam M, Al Mahmud MA, Rahman MT, et al. Determination of green spots (trees) for google satellite images using MATLAB. Proc Comput Sci. 2020;171:1634–41.10.1016/j.procs.2020.04.175Search in Google Scholar

[24] Narjim S, Al Mamun A, Kundu D. Diagnosis of acute lymphoblastic leukemia from microscopic image of peripheral blood smear using image processing technique. International Conference on Cyber Security and Computer Science. vol. 325, 2020. p. 515–26.10.1007/978-3-030-52856-0_41Search in Google Scholar

[25] Nikitin EA, Pavkin DY, Izmailov AY, Aksenov AG. 2022. Assessing the homogeneity of forage mixtures using an RGB camera as exemplified by cattle rations. Appl Sci. 2022;12(7):3230.10.3390/app12073230Search in Google Scholar

[26] Woebbecke DM, Meyer GE, Von Bargen K, Mortensen DA. Color indices for weed identification under various soil, residue, and lighting conditions. Trans ASAE. 1995;38(1):259–69.10.13031/2013.27838Search in Google Scholar

[27] Meyer GE, Neto JC. Verification of color vegetation indices for automated crop imaging applications[J]. Comput Electron Agric. 2008;63(2):282–93.10.1016/j.compag.2008.03.009Search in Google Scholar

[28] Louhaichi M, Borman MM, Johnson DE. Spatially located platform and aerial photography for documentation of grazing impacts on wheat. Geocarto Int. 2001;16(1):65–70.10.1080/10106040108542184Search in Google Scholar

[29] Bendig J, Yu K, Aasen H, Bolten A, Bennertz S, Broscheit J, et al. Combining UAV-based plant height from crop surface models, visible, and near infrared vegetation indices for biomass monitoring in barley. Int J Appl Earth Obs Geoinf. 2015;39:79–87.10.1016/j.jag.2015.02.012Search in Google Scholar

[30] Ridler TW, Calvard S. Picture thresholding using an iterative selection method. IEEE Trans Syst Man, Cybern. 1978;8(8):630–2.10.1109/TSMC.1978.4310039Search in Google Scholar

[31] Sekertekin A. Potential of global thresholding methods for the identification of surface water resources using Sentinel-2 satellite imagery and normalized difference water index. J Appl Remote Sens 2019;13(4):044507.10.1117/1.JRS.13.044507Search in Google Scholar

[32] Yuan L, Yan P, Han W, Huang Y, Wang B, Zhang J, et al. Detection of anthracnose in tea plants based on hyperspectral imaging. Comput Electron Agric. 2019;167:105039.10.1016/j.compag.2019.105039Search in Google Scholar

[33] Zack GW, Rogers WE, Latt SA. Automatic measurement of sister chromatid exchange frequency. J Histochem Cytochem. 1977;25(7):741–53.10.1177/25.7.70454Search in Google Scholar PubMed

[34] Coy A, Rankine D, Taylor M, Nielsen D, Cohen J. 2016. Increasing the accuracy and automation of fractional vegetation cover estimation from digital photographs. Remote Sens. 2016;8(7):474.10.3390/rs8070474Search in Google Scholar

Received: 2024-03-18

Revised: 2024-06-06

Accepted: 2024-06-11

Published Online: 2024-08-30

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

Articles in the same Issue

https://doi.org/10.1515/geo-2022-0661

Keywords for this article

unmanned aerial vehicle; fractional vegetation cover; vegetation indices; image segmentation; vegetation recognition

Creative Commons

BY 4.0