Vegetation coverage automatic extraction method for field camera survey

Abstract

The invention discloses an automatic vegetation coverage extraction method for field camera survey. The method comprises the following steps: inputting an RGB electronic photo to be processed; calculating the overgreen index and the color saturation of the RGB electronic photo; determining a threshold value of over-green index segmentation vegetation pixel by pixel by using the color saturation; identifying a vegetation pixel based on the over-green index and a pixel-by-pixel threshold value; and counting the proportion of the number of vegetation pixels in the RGB electronic picture to the total number of pixels, thereby obtaining the vegetation coverage. According to the method, the optimal threshold value of ExG segmented vegetation can be determined pixel by pixel for RGB electronic photos imaged under different illumination conditions and ground feature scenes, so that high-precision automatic extraction of the vegetation coverage FVC is realized, and vegetation coverage data of a quadrat scale can be quickly and accurately acquired during field investigation; the method is simple in principle, easy to implement, high in adaptive capacity, high in automation degree and wide in application range, and has wide application value in the fields of ecology, geography, remote sensing and the like.

Description

Technical Field

[0001] The present invention relates to the technical fields of remote sensing measurement and image processing, and computer vision technology, and particularly relates to an automatic extraction method for vegetation coverage oriented to field camera surveys using remote sensing image threshold segmentation. Background Art

[0002] Fractional Vegetation Cover (FVC) is an important ecological index describing the surface vegetation coverage state, and is widely used in fields such as carbon cycle, ecological assessment, and soil nutrient cycle. Ground measurement is an important means to obtain accurate FVC data. Among them, the photographic method has gradually become an important FVC measurement method in field surveys due to its simplicity and high efficiency. The photographic method obtains quantitative information on vegetation coverage by collecting visible light true color electronic photos and using computer vision technology to identify and segment green vegetation areas. In the vegetation recognition method based on RGB electronic photos, the Excess Green Index (ExG) is a commonly used green vegetation recognition index and is widely used because it can efficiently distinguish vegetation and non-vegetation areas. However, the selection of the optimal segmentation threshold of ExG is strongly dependent on imaging conditions and image backgrounds. Especially in the case of complex lighting imaging conditions and insufficient background contrast, its recognition accuracy significantly decreases, limiting its automated application in actual scenarios.

[0003] Currently, common image threshold segmentation methods include global threshold methods and local threshold methods. Global threshold methods are applicable to images with uniform lighting and obvious contrast between the foreground and the background. Common global threshold segmentation methods include automatic threshold determination methods such as the maximum entropy threshold method and the Otsu method. The maximum entropy threshold method performs well in low-density vegetation areas, and the Otsu method has advantages in maximizing the between-class variance segmentation. However, no matter which global threshold method is used, its recognition accuracy is limited in imaging scenarios with strong lighting changes. Lighting changes make the optimal segmentation thresholds of the ExG threshold different in strong light and weak light areas, and a single global threshold method is difficult to meet the segmentation requirements of different lighting areas. In contrast, local threshold methods show stronger robustness to images with uneven lighting due to their ability to refine local features. Summary of the Invention

[0004] In view of the deficiency that the existing greenness index ExG segmentation threshold for extracting vegetation is vulnerable to imaging illumination conditions, the present invention proposes an automated method for extracting vegetation coverage for field camera surveys. It determines the optimal threshold for ExG to segment vegetation for each pixel using the color saturation S, thereby realizing the automated extraction of vegetation coverage from RGB electronic photos. This method can determine the optimal threshold for ExG to segment vegetation for each pixel of RGB electronic photos taken under different illumination conditions and ground object scenes, so as to achieve high-precision automatic extraction of FVC, facilitating the rapid and accurate acquisition of vegetation coverage data at the quadrat scale during field surveys. Its principle is simple, easy to implement, has strong adaptability, high automation, and a wide range of applications.

[0005] To achieve this purpose, the automated method for extracting vegetation coverage for field camera surveys of the present invention includes the following steps:

[0006] Input the RGB electronic photo to be processed;

[0007] Calculate the greenness index and color saturation of the RGB electronic photo;

[0008] Use the color saturation to determine the threshold for the greenness index to segment vegetation for each pixel;

[0009] Identify vegetation pixels based on the greenness index and the pixel-by-pixel threshold;

[0010] Count the proportion of vegetation pixels in the total number of pixels in the RGB electronic photo to obtain the vegetation coverage.

[0011] Optionally, in the step of using the color saturation to determine the threshold for the greenness index to segment vegetation for each pixel:

[0012] The threshold for the greenness index to segment vegetation is determined only by the color saturation for each pixel;

[0013] The threshold for the greenness index to segment vegetation is a regression fitting function of the color saturation.

[0014] Optionally, the numerical ranges of the red, green, and blue bands of the RGB electronic photo are 0 - 255, which are byte-type data.

[0015] Optionally, in the step of calculating the greenness index and color saturation of the RGB electronic photo: Calculate the greenness index and color saturation for each pixel.

[0016] Optionally, in the step of using the color saturation to determine the threshold for the greenness index to segment vegetation for each pixel, the adaptive model of the greenness index threshold used is obtained by training with RGB electronic photos collected under different illumination conditions, vegetation coverage situations, and complex land type combination scenes.

[0017] Optionally, in the training step, it includes:

[0018] Based on these data, interpret the vegetation and non-vegetation attributes pixel by pixel;

[0019] The relationship between the threshold T of the Excess Green Index and the color saturation is fitted by a regression method;

[0020] The adaptive model of the threshold of the Excess Green Index is:

[0021]

[0022] Optionally, it further includes the step of calculating the segmentation threshold of the Excess Green Index for each pixel by using the color saturation of each pixel of the RGB electronic photo to be processed and combining with the adaptive model of the threshold of the Excess Green Index.

[0023] Optionally, in the step of identifying vegetation pixels based on the Excess Green Index and the pixel-by-pixel threshold, for each pixel of the RGB electronic photo to be processed, compare the calculated Excess Green Index with the threshold to determine whether the pixel is vegetation.

[0024] Optionally, the calculation formula of the color saturation S is:

[0025]

[0026] In the formula, , and are the normalized values of the red, green, and blue bands in the RGB electronic photo respectively, is a minimum value used to avoid the denominator being zero.

[0027] Optionally, for each pixel of the RGB electronic photo, compare its Excess Green Index ExG denoted as ExG(x, y) with the adaptive threshold T to determine whether the pixel attribute is vegetation or non-vegetation, and the judgment condition is:

[0028]

[0029] Wherein, is the pixel classification result, 1 represents vegetation, and 0 represents non-vegetation.

[0030] The present invention has the following characteristics:

[0031] (1) Strong adaptability, applicable to RGB electronic photos with different lighting conditions and ground object scenes;

[0032] (2) Simple principle and easy to implement. The pixel-by-pixel optimal segmentation threshold of the Excess Green Index ExG is only determined by the color saturation S;

[0033] (3) High degree of automation. Once the ExG adaptive threshold model is established based on training samples, high-precision extraction of vegetation coverage can be automatically achieved in the study area.

[0034] (4) Wide range of applications. It can be widely applied in the fields of ecology, geography, remote sensing, etc. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 It is a flowchart of the automatic extraction method of vegetation coverage for field camera surveys according to the present invention.

[0036] Figure 2 It is an RGB digital photo of the automatic extraction method of vegetation coverage for field camera surveys according to the present invention with an unmanned aerial vehicle as an example.

[0037] Figure 3 It is an excess green index ExG feature image of the automatic extraction method of vegetation coverage for field camera surveys according to the present invention.

[0038] Figure 4 It is a color saturation S feature image of the automatic extraction method of vegetation coverage for field camera surveys according to the present invention.

[0039] Figure 5 It is an ExG-S scatter plot and an ExG adaptive threshold model of the automatic extraction method of vegetation coverage for field camera surveys according to the present invention.

[0040] Figure 6 It is a vegetation recognition result image of the automatic extraction method of vegetation coverage for field camera surveys according to the present invention, where the black area represents vegetation and the white area represents non-vegetation. DETAILED DESCRIPTION OF THE INVENTION

[0041] The inventor found in the study of threshold segmentation of remote sensing images that: color saturation (Saturation, S), as an index to describe the vividness of colors, can reflect the visual characteristics of vegetation under different lighting conditions. Therefore, S can be used to indicate the different lighting conditions of vegetation in RGB digital photos. Both the S value and the ExG value show co-variation with the lighting conditions: under strong light conditions, the vegetation color is vivid, the S value and the ExG value are higher, and the vegetation segmentation threshold needs to be increased accordingly; under weak light conditions, the vegetation color is dull, the S value and the ExG value are lower, and the vegetation segmentation threshold should be appropriately reduced. Based on this, the S value can be used as a key factor to determine the optimal ExG segmentation threshold, which can be used to overcome the influence of lighting changes in the image on vegetation recognition, so as to achieve high-precision and automatic extraction of FVC based on RGB digital photos.

[0042] Therefore, aiming at the deficiency that the existing greenness index ExG segmentation threshold is vulnerable to imaging illumination conditions when extracting vegetation, an automated method for extracting vegetation coverage for field camera surveys is proposed, including the steps:

[0043] Input the RGB electronic photo to be processed;

[0044] Calculate the greenness index and color saturation of the RGB electronic photo;

[0045] Use the color saturation to determine the threshold for segmenting vegetation by the greenness index pixel by pixel;

[0046] Identify vegetation pixels based on the greenness index and the pixel-by-pixel threshold;

[0047] Count the proportion of the number of vegetation pixels in the RGB electronic photo to the total number of pixels, and thus obtain the vegetation coverage.

[0048] In one embodiment, in the step of using the color saturation to determine the threshold for segmenting vegetation by the greenness index pixel by pixel:

[0049] The threshold for segmenting vegetation by the greenness index is determined only by the color saturation pixel by pixel;

[0050] The threshold for segmenting vegetation by the greenness index is a regression fitting function of the color saturation.

[0051] In one embodiment, the numerical ranges of the red, green, and blue bands of the RGB electronic photo are 0 - 255, which are byte-type data.

[0052] In one embodiment, in the step of calculating the greenness index and color saturation of the RGB electronic photo: Calculate the greenness index and color saturation of each pixel.

[0053] In one embodiment, in the step of using the color saturation to determine the threshold for segmenting vegetation by the greenness index pixel by pixel, the adaptive model of the greenness index threshold used is obtained by training with RGB electronic photos collected under different illumination conditions, vegetation coverage situations, and complex land type combination scenarios.

[0054] In one embodiment, in the training step, it includes:

[0055] Based on these data, interpret the vegetation and non-vegetation attributes pixel by pixel;

[0056] Fit the relationship between the threshold T of the greenness index and the color saturation through a regression method;

[0057] The adaptive model of the greenness index threshold is:

[0058]

[0059] In one embodiment, it further includes the step of calculating the segmentation threshold of the excess green index for each pixel by using the color saturation of each pixel of the RGB electronic photo to be processed and combining with the threshold adaptive model of the excess green index.

[0060] In one embodiment, in the step of identifying vegetation pixels with the excess green index and per-pixel threshold, for each pixel of the RGB electronic photo to be processed, the calculated excess green index is compared with the threshold to determine whether the pixel is vegetation.

[0061] In one embodiment, the calculation formula of the color saturation S is:

[0062]

[0063] In the formula, , and are the normalized values of the red, green, and blue bands in the RGB electronic photo respectively, is a minimum value used to avoid the denominator being zero.

[0064] In one embodiment, for each pixel of the RGB electronic photo, its excess green index ExG denoted as ExG(x, y) is compared with the adaptive threshold T to determine whether the pixel attribute is vegetation or non-vegetation, and the judgment condition is:

[0065]

[0066] Wherein, is the pixel classification result, 1 represents vegetation, and 0 represents non-vegetation.

[0067] To further illustrate the technical implementation scheme of the present invention, the following takes a drone RGB electronic photo as an example and in combination with the accompanying drawings to introduce in detail the technical implementation process of an automatic extraction method for vegetation coverage degree for field camera surveys. The overall technical process is as Figure 1 shown. The specific implementation steps are as follows:

[0068] Previously, use the field camera of the drone to obtain the RGB electronic photo of the area to be surveyed.

[0069] Input the RGB electronic photo to be processed.

[0070] In this step, input a drone RGB electronic photo to be processed as Figure 2 shown. In this embodiment, the spatial resolution of this image is 5 cm, and there are a total of 5472 columns × 3648 rows of pixels, and the value ranges of the red, green, and blue bands are all 0 - 255, byte-type data.

[0071] Next, calculate the excess green index ExG and the color saturation S of the RGB electronic photo.

[0072] Specifically, according to the RGB electronic photo to be processed, calculate the excess green index ExG and the color saturation S of each pixel therein.

[0073] For each pixel of the RGB electronic photo, calculate the excess green index ExG and its corresponding color saturation S.

[0074] The formula for calculating ExG is:

[0075] , , (1) , , (2) (3)

[0076] In the formula, , and are the pixel values of the red, green, and blue bands in the RGB electronic photo respectively, , and are the maximum pixel values of the red, green, and blue bands respectively, , and are the normalized pixel values of the red, green, and blue bands respectively.

[0077] The formula for calculating S is:

[0078] (4)

[0079] In the formula, , and are the normalized values of the red, green, and blue bands in the RGB electronic photo (Equation 1) respectively, is a minimum value used to avoid the case where the denominator is zero.

[0080] The ExG image and the S feature image obtained in this step are as shown in Figure 3 and Figure 4 shown.

[0081] Next, use the color saturation S of each pixel to determine the threshold for segmenting vegetation with the excess green index ExG pixel by pixel, and the threshold is the optimal threshold.

[0082] Use the color saturation S of each pixel of the RGB electronic photo, substitute it into the ExG adaptive threshold model, and determine the optimal ExG threshold for segmenting vegetation pixel by pixel. The ExG adaptive threshold model is:

[0083] (5)

[0084] In the formula, is the adaptive threshold of each pixel of ExG, is the saturation value of the current pixel, is a parameterized model driven by training data. The input parameter of this model is only S, and the model representation can be a linear or non-linear function. Typical forms include, but are not limited to, linear models, quadratic polynomial models, and exponential models. The model parameters are determined by regression fitting of the training set data in the study area.

[0085] In this embodiment, the threshold adaptive model of the greenness index used is trained by 50 RGB electronic photos (resolution 1-5 cm) collected under different lighting conditions, vegetation coverage, and complex land cover combination scenarios. First, calculate the greenness index ExG and color saturation S of each pixel. Based on these data, interpret the vegetation and non-vegetation attributes of each pixel one by one, and through regression methods (such as support vector machine regression SVM), use the greenness index ExG and color saturation S to fit the relationship between the optimal threshold T of the greenness index and the color saturation S. The relationship between ExG and S is visualized through scatter Figure 5 plots, and the threshold adaptive model of the greenness index in this embodiment is obtained:

[0086]

[0087] Using the color saturation S value of each pixel of the RGB electronic photo to be processed, combine the above threshold adaptive model of the greenness index to calculate the optimal segmentation threshold T of the greenness index ExG of each pixel.

[0088] Next, identify vegetation pixels based on the greenness index ExG and the pixel-by-pixel threshold.

[0089] For each pixel of the RGB electronic photo, compare its ExG value (denoted as ExG(x, y)) with the adaptive threshold T to determine whether the pixel attribute is vegetation or non-vegetation. The judgment condition is:

[0090] (6)

[0091] Where, is the pixel classification result, 1 represents vegetation, and 0 represents non-vegetation.

[0092] For each pixel of the RGB electronic photo to be processed, compare the calculated ExG value with the threshold T to determine whether the pixel is vegetation. The final vegetation recognition result is as Figure 6 shown, where the black area represents vegetation and the white area represents non-vegetation.

[0093] Finally, the proportion of vegetation pixels in the total pixels of the RGB digital photo is counted, and the fractional vegetation cover (FVC) is obtained therefrom.

[0094] The proportion of vegetation pixels to the total pixels is counted, and the fractional vegetation cover (FVC) is calculated. The calculation formula of FVC is as follows:

[0095] (7)

[0096] In the formula, is the number of pixels identified as vegetation in the RGB digital photo, is the total number of pixels.

[0097] The proportion of vegetation pixels to the total pixels in the RGB digital photo is counted, and the fractional vegetation cover (FVC) is calculated. In this case, the fractional vegetation cover of the RGB digital photo is calculated to be 0.68 (i.e., 68%), and the recognition accuracy is 91%.

[0098] The present invention realizes the automatic extraction of the fractional vegetation cover of RGB digital photos. This method can determine the optimal threshold for ExG segmentation of vegetation pixel by pixel for RGB digital photos taken under different lighting conditions and ground object scenes, so as to realize the high-precision automatic extraction of FVC, which is convenient for quickly and accurately obtaining the fractional vegetation cover data at the quadrat scale during field surveys. Its principle is simple, easy to implement, has strong adaptability, high automation, and wide application range. It has the following characteristics:

[0099] (1) Strong adaptability, applicable to RGB digital photos taken under different lighting conditions and ground object scenes;

[0100] (2) Simple principle and easy to implement. The optimal segmentation threshold of the excess green index (ExG) for each pixel is only determined by the color saturation S;

[0101] (3) High automation. Once the ExG adaptive threshold model is established based on training samples, high-precision extraction of the fractional vegetation cover can be automatically realized in the study area;

[0102] (4) Wide application range, can be widely applied to the fields of ecology, geography, remote sensing, etc.

[0103] As described above, it is only a specific implementation manner of the present invention. The ExG threshold adaptive model is not the only fixed one. The model representation forms include but are not limited to the unary linear model, and the model parameters will also vary with the change of the training data set. Therefore, the protection scope of the present invention should not be limited to a specific regression method or parameter range. Any reasonable changes or substitutions made by those familiar with the technical field based on the technical content disclosed by the present invention should be covered within the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims.

Claims

1. An automatic extraction method of vegetation coverage for field camera surveys, characterized in that: The following steps are involved: Input the RGB electronic photo to be processed; Calculate the over-green index and color saturation of RGB electronic photos; The threshold of overgreen index segmentation of vegetation is determined pixel by pixel using color saturation. Identify vegetation pixels based on overgreening index and pixel-by-pixel threshold; The vegetation coverage was obtained by counting the proportion of vegetation pixels to the total number of pixels in the RGB electronic photo.

2. The method for automatically extracting vegetation coverage for field camera survey according to claim 1, characterized in that: In the threshold step of using color saturation to determine the overgreen index pixel by pixel to segment vegetation: The threshold for segmenting vegetation by the overgreen index is determined pixel by pixel only by color saturation; The threshold for segmenting vegetation using the overgreen index is a regression fitting function of color saturation.

3. The method for automatically extracting vegetation coverage for field camera survey according to claim 1, characterized in that: The value range of each band of red, green and blue of the RGB electronic photo is 0-255, which is byte data.

4. The method for automatically extracting vegetation coverage for field camera survey according to claim 1, characterized in that: In the step of calculating the over-green index and color saturation of the RGB electronic photograph: the over-green index and color saturation of each pixel are calculated.

5. The method for automatically extracting vegetation coverage for field camera survey according to claim 1, characterized in that: In the step of using color saturation to determine the threshold of the overgreen index for segmenting vegetation pixel by pixel, the overgreen index threshold adaptive model used is obtained by training RGB electronic photos collected under different lighting conditions, vegetation coverage and complex land type combination scenes.

6. The method for automatically extracting vegetation coverage for field camera survey according to claim 5, characterized in that: In the training step, include: Based on these data, vegetation and non-vegetation attributes are interpreted pixel by pixel; The relationship between the threshold value T of the over-green index and the color saturation S is fitted by regression method; The threshold adaptive model of the overgreen index is: 。 7. The method for automatically extracting vegetation coverage for field camera survey according to claim 6, characterized in that: The method also includes the steps of: using the color saturation of each pixel of the RGB electronic photo to be processed and combining the threshold adaptive model of the over-green index to calculate the segmentation threshold of the over-green index of each pixel.

8. The method for automatically extracting vegetation coverage for field camera survey according to claim 6, characterized in that: In the step of identifying vegetation pixels based on the overgreen index and the pixel-by-pixel threshold, for each pixel of the RGB electronic photo to be processed, the calculated overgreen index is compared with the threshold to determine whether the pixel is vegetation.

9. The method for automatically extracting vegetation coverage for field camera survey according to claim 6, characterized in that: The calculation formula of color saturation S is: ， In the formula, , and They are the normalized values ​​of the red, green and blue bands in the RGB electronic photo, is a minimum value used to avoid the denominator being zero.

10. The method for automatically extracting vegetation coverage for field camera survey according to claim 6, characterized in that: For each pixel of the RGB electronic photo, its overgreen index ExG is recorded as ExG(x, y) and compared with the adaptive threshold T to determine whether the pixel attribute is vegetation or non-vegetation. The judgment condition is: ， in, is the pixel classification result, 1 represents vegetation and 0 represents non-vegetation.

Images