An image recognition-based organic fertilizer field-returned pasture growth monitoring method and system

Abstract

The application discloses an organic fertilizer returning field pasture growth monitoring method and system based on image recognition, and particularly relates to the field of image recognition processing, and is used for solving the problem that the existing pasture growth monitoring method cannot effectively distinguish the reflectivity difference caused by the surface geometric structure and the reflectivity change caused by the pasture biomass difference from the image level; the difference between the concentration of the gradient direction of the pixel in the orthographic image in the organic fertilizer strip application direction and the orthogonal direction is calculated to determine the light and dark alternating texture, and then the edge gradient width of the gray profile is used to distinguish the texture causes; the skewness coefficient and the kurtosis coefficient of the pixel belonging to the micro-terrain reflection difference are used to separate the reflectivity components based on the brightness probability density distribution to obtain a corrected image; and the vegetation index is extracted from the corrected image to construct a growth spatial distribution map.

Description

Technical Field

[0001] This invention relates to the field of image recognition and processing technology, and more specifically, to a method and system for monitoring the growth of forage grasses in fields using organic fertilizer returned to the soil based on image recognition. Background Technology

[0002] In mountainous agricultural and pastoral areas, the crop-livestock cycle model returns the organic fertilizer from fermented bedding to pasture fields to replenish soil organic matter and increase pasture yield. To monitor pasture growth after organic fertilizer application, drones equipped with visible light or multispectral cameras are used to take aerial photographs of pasture plots. Image recognition methods are then used to extract growth indicators such as vegetation cover and canopy greenness from the stitched orthophotos, which is a commonly used non-contact monitoring method. Due to limitations in mountainous terrain and agronomic practices, organic fertilizer is often applied in strips along slopes or contour lines, creating regular micro-topographic undulations between the fertilized furrows and unfertilized zones.

[0003] During this monitoring process, the surface micro-topographic texture formed by the strip application of organic fertilizer will result in alternating light and dark bands along the fertilization direction in the orthophoto. Since the spatial orientation of the alternating light and dark bands is completely consistent with the layout of organic fertilizer application, existing methods cannot effectively distinguish between the reflectance differences caused by the surface geometry and the reflectance changes caused by the biomass differences of the forage grass itself when extracting the spatial distribution characteristics of forage grass growth. This results in distorted spectral characterization of some areas in the stitched image, ultimately causing a deviation in the interpretation of the spatial distribution of forage grass growth and affecting the accuracy of the assessment of the effect of organic fertilizer application. Summary of the Invention

[0004] In order to overcome the above-mentioned defects of the prior art, the present invention provides an image recognition-based organic fertilizer returning to the field forage growth monitoring system to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: A method for monitoring the growth of forage grasses after organic fertilizer is returned to the field, based on image recognition, includes: S1: Acquire orthophotos of pasture plots and simultaneously record the direction of organic fertilizer strip application; S2: Calculate the first concentration of the pixel gradient direction in the orthophoto image within the direction interval where the organic fertilizer strip application direction is located, and the second concentration within the direction interval orthogonal to the organic fertilizer strip application direction. When the difference between the first concentration and the second concentration exceeds the preset concentration difference threshold, it is determined that there is a light and dark alternating texture and S3 is executed; otherwise, the orthophoto image is used as the analysis image and S5 is executed. S3: Extract grayscale profiles along the direction of the alternating light and dark texture, calculate the edge gradient width of the grayscale profiles, and if the edge gradient width is less than the preset width threshold, determine that the alternating light and dark texture belongs to micro-topographical reflection difference and execute S4; otherwise, use the orthophoto as the analysis image and execute S5. S4: Establish the luminance probability density distribution for the pixels containing alternating light and dark textures, and separate the reflectance components of each pixel based on the skewness coefficient and kurtosis coefficient of the luminance probability density distribution to obtain a corrected image that eliminates micro-topographic reflection differences as the analysis image. S5: Extract vegetation indices reflecting the growth status of pasture canopy from analyzed images; S6: Construct a spatial distribution map of the growth of pasture plots using vegetation indices.

[0006] Further, S1 includes: under cloudy uniform diffused light conditions, acquiring multi-view sequence images from above the pasture plot at a vertically downward shooting angle and a preset spatial resolution; performing motion recovery structure processing on the acquired multi-view sequence images to generate a three-dimensional point cloud of the pasture plot; performing orthorectification on the three-dimensional point cloud of the pasture plot to obtain an orthophoto of the pasture plot; and recording the direction of organic fertilizer strip application while acquiring the orthophoto of the pasture plot.

[0007] Further, S2 includes: using the Sobel operator to calculate the gradient magnitude and gradient direction of each pixel in the orthophoto; dividing the direction interval with the organic fertilizer strip application direction as the reference direction into a parallel direction interval and an orthogonal direction interval; statistically analyzing the gradient direction distribution histogram in the parallel direction interval and calculating the first concentration; statistically analyzing the gradient direction distribution histogram in the orthogonal direction interval and calculating the second concentration; when the difference between the first concentration and the second concentration exceeds a preset concentration difference threshold, it is determined that there is an alternating light and dark texture in the orthophoto.

[0008] Further, S3 includes: determining the texture direction from the area where the alternating light and dark textures are located in the orthophoto; extracting grayscale profiles one by one along the direction perpendicular to the texture direction and recording the pixel grayscale sequence and pixel spatial coordinates of each grayscale profile; calculating the first-order difference of the pixel grayscale sequence of each grayscale profile to obtain the gradient sequence; locating the pixel position corresponding to the maximum gradient value in the gradient sequence as the position of the light-dark transition edge; expanding outwards from the position of the light-dark transition edge to both sides at a preset ratio to determine the edge transition area; calculating the full width at half maximum of the gradient sequence in the edge transition area as the edge gradient width; comparing the average edge gradient width of multiple grayscale profiles with a preset width threshold; if the average value is less than the preset width threshold, determining that the alternating light and dark texture belongs to micro-topographic reflection difference and executing S4; otherwise, directly using the orthophoto as the analysis image and executing S5.

[0009] Further, determining the texture direction includes: obtaining the gradient direction of each pixel in the area where the alternating light and dark texture is located; using the first concentration and the second concentration calculated in S2 to determine the gradient direction dominance distribution interval; extracting the median gradient direction within the gradient direction dominance distribution interval as the main texture direction; and using the main texture direction as the texture direction of the alternating light and dark texture.

[0010] Further, S4 includes: establishing a brightness probability density distribution curve for each pixel containing alternating light and dark textures that are determined to belong to micro-topographic reflection differences; calculating the skewness coefficient and kurtosis coefficient of the brightness probability density distribution curve; distinguishing between light-facing slope pixels and shadow-facing slope pixels by the sign of the skewness coefficient; quantifying the degree of scattering asymmetry by the absolute value of the skewness coefficient; characterizing the concentration of scattering patterns by the deviation of the kurtosis coefficient from the kurtosis value of the normal distribution; constructing anisotropic scattering component weights by fusing the degree of scattering asymmetry and the concentration of scattering patterns; using the anisotropic scattering component weights to strip the anisotropic scattering components and retain the Lambertian reflection components from the original brightness values ​​of the pixels; merging the stripped Lambertian reflection components to obtain a corrected image that eliminates micro-topographic reflection differences; and using the corrected image as the analysis image.

[0011] Furthermore, S5 includes: acquiring the spectral reflectance of the pasture canopy cover pixels in the analysis image; calculating the ratio of the red edge band and the near-infrared band of the spectral reflectance to obtain the normalized vegetation index; calculating the normalized difference between the green light band and the red light band of the spectral reflectance to obtain the visible light atmospheric resistance index; and multiplying the normalized vegetation index and the visible light atmospheric resistance index to generate a vegetation index reflecting the growth status of the pasture canopy.

[0012] Furthermore, generating a vegetation index reflecting the growth status of the pasture canopy includes: acquiring corrected pixel areas identified as belonging to micro-topographic reflectance differences and non-differential pixel areas not identified as belonging to micro-topographic reflectance differences in the analyzed image; calculating the first product of the normalized vegetation index and the visible light resistance index of the corrected pixel areas; calculating the second product of the normalized vegetation index and the visible light resistance index of the non-differential pixel areas; assigning a correction weight coefficient determined by the weight of the anisotropic scattering component in S4 to the first product; assigning a unit weight coefficient to the second product; and combining the weighted first product and the weighted second product to generate a vegetation index reflecting the growth status of the pasture canopy.

[0013] Furthermore, S6 includes: acquiring a spatial distribution array of vegetation indices; constructing a spatial reference frame for the growth spatial distribution map based on the pixel grid of the orthophoto of the pasture patch; mapping each vegetation index value in the spatial distribution array of vegetation indices pixel by pixel to the corresponding pixel grid position in the spatial reference frame; performing partitioned, graded, and color-coded rendering on the mapped pixel grid to generate a growth spatial distribution map of the pasture patch.

[0014] On the other hand, the present invention provides an image recognition-based system for monitoring the growth of forage grass after organic fertilizer is returned to the field, comprising: The image acquisition module is used to acquire orthophotos of pasture plots and simultaneously record the direction of organic fertilizer strip application; The texture detection module is used to calculate the first concentration of the pixel gradient direction in the orthophoto image within the direction interval where the organic fertilizer strip application direction is located, and the second concentration within the direction interval orthogonal to the organic fertilizer strip application direction. When the difference between the first concentration and the second concentration exceeds the preset concentration difference threshold, it is determined that there is a texture with alternating light and dark and the cause discrimination module is executed. Otherwise, the orthophoto image is used as the analysis image and the index extraction module is executed. The cause discrimination module is used to extract grayscale profiles along the direction of the vertical alternating light and dark texture, calculate the edge gradient width of the grayscale profile, and if the edge gradient width is less than the preset width threshold, it is determined that the alternating light and dark texture belongs to micro-topographic reflection difference and the reflection decoupling module is executed; otherwise, the orthophoto is used as the analysis image and the exponential extraction module is executed. The reflection decoupling module is used to establish the brightness probability density distribution of the pixels containing alternating light and dark textures. Based on the skewness coefficient and kurtosis coefficient of the brightness probability density distribution, the reflectivity components of each pixel are separated to obtain a corrected image that eliminates micro-topographic reflection differences as the analysis image. The index extraction module is used to extract vegetation indices reflecting the growth status of pasture canopies from the analyzed images; The distribution construction module is used to construct a spatial distribution map of the growth of pasture plots based on vegetation indices.

[0015] Compared with the prior art, the present invention has the following beneficial effects: 1. The existence of alternating light and dark textures is determined by detecting the difference in the concentration of pixel gradient directions in orthophotos within the intervals of the direction where organic fertilizer strips are applied and the intervals of orthogonal directions. Texture recognition is performed by utilizing the physical characteristic that micro-topographic textures have strong directional selectivity in gradient direction space. From the perspective of directional specificity, regular light and dark stripes caused by strip application micro-topography and random textures caused by differences in the biomass of pasture itself are distinguished. After determining the existence of alternating light and dark textures, grayscale profiles are further extracted along the direction perpendicular to the texture direction and the edge gradient width is calculated. Based on the physical difference of sharp transitions at the edges of micro-topographic shadows and smooth transitions at the edges of biomass differences, the cause of alternating light and dark textures is automatically determined, so that pseudo-textures caused by differences in micro-topographic reflection are accurately identified and not misjudged as differences in pasture growth.

[0016] 2. After confirming that the alternating light and dark textures are due to micro-topographic reflection differences, a brightness probability density distribution is established for relevant pixels. Skewness coefficients are used to distinguish between pixels on the sunlit slope and pixels on the shaded slope and to quantify the degree of scattering asymmetry. Kurtosis coefficients are used to characterize the concentration of scattering patterns in single scattering and multiple scattering. Anisotropic scattering component weights are constructed by fusing the degree of scattering asymmetry and the concentration of scattering patterns. Anisotropic scattering components are stripped from pixels and Lambertian reflection components are retained to achieve adaptive correction that eliminates micro-topographic reflection differences. The corrected image is used as an analysis image for vegetation index extraction and growth spatial distribution map construction. All processing is completed at the image level without relying on additional sensor data or field measured parameters. Attached Figure Description

[0017] Figure 1 This is a flowchart of a method for monitoring the growth of forage grass by returning organic fertilizer to the field based on image recognition, according to the present invention. Figure 2 This is a schematic diagram of the structure of an image recognition-based organic fertilizer return to field pasture growth monitoring system according to the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] Example 1: Figure 1 This invention presents a method for monitoring the growth of forage grasses when organic fertilizer is returned to the field, based on image recognition, comprising: S1: Acquire orthophotos of pasture plots and simultaneously record the direction of organic fertilizer strip application; S2: Calculate the first concentration of the pixel gradient direction in the orthophoto image within the direction interval where the organic fertilizer strip application direction is located, and the second concentration within the direction interval orthogonal to the organic fertilizer strip application direction. When the difference between the first concentration and the second concentration exceeds the preset concentration difference threshold, it is determined that there is a light and dark alternating texture and S3 is executed; otherwise, the orthophoto image is used as the analysis image and S5 is executed. S3: Extract grayscale profiles along the direction of the alternating light and dark texture, calculate the edge gradient width of the grayscale profiles, and if the edge gradient width is less than the preset width threshold, determine that the alternating light and dark texture belongs to micro-topographical reflection difference and execute S4; otherwise, use the orthophoto as the analysis image and execute S5. S4: Establish the luminance probability density distribution for the pixels containing alternating light and dark textures, and separate the reflectance components of each pixel based on the skewness coefficient and kurtosis coefficient of the luminance probability density distribution to obtain a corrected image that eliminates micro-topographic reflection differences as the analysis image. S5: Extract vegetation indices reflecting the growth status of pasture canopy from analyzed images; S6: Construct a spatial distribution map of the growth of pasture plots using vegetation indices.

[0020] In the specific implementation of the embodiment corresponding to S1, under cloudy uniform diffused light conditions, multi-view image sequences are acquired from above the pasture plots at a vertically downward shooting angle and a preset spatial resolution. Cloudy uniform diffused light conditions refer to a lighting environment where the sun is completely blocked by clouds, with atmospheric diffused light uniformly illuminating the pasture plots, cloud cover exceeding 80%, and no shadows cast by direct sunlight on the pasture plot surface. The preset spatial resolution is determined based on the minimum leaf width of the pasture canopy; for example, it is set to half the minimum leaf width of the pasture canopy to ensure that individual pixels in subsequent orthophotos can distinguish the leaf-level structure of the pasture canopy. When acquiring multi-view image sequences, a UAV equipped with a visible light camera flies along a predetermined route covering the entire pasture plot, with a forward overlap of no less than 70% and a lateral overlap of no less than 70%. The optical axis of the visible light camera is always perpendicular to the horizontal plane to ensure sufficient redundancy information between adjacent images in the multi-view image sequence for motion reconstruction processing.

[0021] Structure-of-motion (SOR) processing is performed on the acquired multi-view image sequence to generate a 3D point cloud of grassland patches. SOR processing extracts scale-invariant feature points from each image in the multi-view sequence. Feature matching is performed between different images to obtain matching feature point pairs. Using the disparity relationship of the feature points in the matching pairs, bundle adjustment is used to simultaneously calculate the spatial position parameters, attitude parameters, and 3D spatial coordinates of the feature points corresponding to each image at the time of capture. Bundle adjustment aims to minimize the sum of squares of the reprojection errors of all feature points across all images, iteratively solving for the optimal estimates of the spatial position parameters, attitude parameters, and 3D spatial coordinates of the feature points. After bundle adjustment, a multi-view stereo matching algorithm is used, based on the calculated spatial position and attitude parameters of the visible light camera, to perform pixel-by-pixel dense matching on the multi-view image sequence, generating a dense 3D point cloud of grassland patches. Each point in the dense 3D point cloud contains 3D spatial coordinates and visible light band reflectivity information.

[0022] Orthorectification is performed on the 3D point cloud of pasture blocks to obtain orthorectified images of the pasture blocks. Orthorectification uses the average elevation surface of the pasture blocks as the projection plane. The visible light reflectance information of each point in the dense 3D point cloud is projected onto the projection plane along a direction perpendicular to the projection plane. The projection plane is divided into a regular grid, with each grid serving as a pixel in the orthorectified image. The pixel size is equal to the preset spatial resolution. When multiple points in the dense 3D point cloud are projected onto the same pixel, the visible light reflectance information corresponding to the point with the highest elevation value among all points projected onto that pixel is taken as the brightness value of that pixel. When no points in the dense 3D point cloud are projected onto a certain pixel, the brightness value of that pixel is filled using distance-weighted interpolation based on the brightness values ​​of adjacent pixels. The weight of the distance-weighted interpolation is the reciprocal of the Euclidean distance between the center of the pixel to be filled and the centers of adjacent pixels. After assigning brightness values ​​to all pixels, the orthorectified image of the pasture blocks is generated.

[0023] While acquiring orthophotos of pasture plots, the direction of organic fertilizer strip application is recorded. The direction of organic fertilizer strip application is the azimuth angle of the fertilizer furrow's extension along the ground surface during strip application within the orthophoto's pixel coordinate system. The direction is recorded as the angle between the positive X-axis of the orthophoto coordinate system and the direction of the fertilizer furrow. For example, when the fertilizer furrow's direction is parallel to the row direction of the orthophoto, the direction is recorded as 0 degrees; when the fertilizer furrow's direction is parallel to the column direction of the orthophoto, it is recorded as 90 degrees. The direction of organic fertilizer strip application is obtained by reading the trajectory data of agricultural machinery during strip application, or by manually marking the start and end coordinates of the fertilizer furrow in the orthophoto and then calculating the direction of the line connecting the start and end coordinates.

[0024] In the specific implementation of the embodiment corresponding to S2, the Sobel operator is used to calculate the gradient magnitude and gradient direction of each pixel in the orthophoto. The Sobel operator includes a horizontal convolution kernel and a vertical convolution kernel. The horizontal convolution kernel is a 3x3 matrix, with the first row containing -1, 0, 1, the second row containing -2, 0, 2, and the third row containing -1, 0, 1. The vertical convolution kernel is also a 3x3 matrix, with the first row containing -1, -2, -1, the second row containing 0, 0, 0, and the third row containing 1, 2, 1. The horizontal gradient value is obtained by convolving the horizontal convolution kernel with a 3×3 neighborhood window centered on the pixel to be calculated in the orthophoto. The vertical gradient value is obtained by convolving the vertical convolution kernel with the same 3×3 neighborhood window. The gradient magnitude G of each pixel is calculated as: Ga = (G x 2 +Gy 2G is obtained by taking the square root of Ga; where Gx represents the horizontal gradient value and Gy represents the vertical gradient value. The gradient direction θ of each pixel is calculated as: θ = arctan(Gy / Gx); where arctan represents the arctangent function, the unit of gradient direction θ is degrees, and the range of gradient direction θ is 0 degrees to 180 degrees. A gradient direction θ of 0 degrees indicates that the pixel brightness increases horizontally, and a gradient direction θ of 90 degrees indicates that the pixel brightness increases vertically. For pixels with a horizontal gradient value Gx of zero, the gradient direction θ is set to 90 degrees.

[0025] The directional interval, with the organic fertilizer strip application direction as the reference direction, is divided into a parallel directional interval and an orthogonal directional interval. The organic fertilizer strip application direction is derived from the organic fertilizer strip application direction recorded in S1, and the unit for the organic fertilizer strip application direction is degrees. The parallel directional interval is defined as the directional interval extending to both sides of the organic fertilizer strip application direction within a preset angular tolerance range. The parallel directional interval includes the organic fertilizer strip application direction and all gradient directions whose deviation from the organic fertilizer strip application direction does not exceed the preset angular tolerance range. The orthogonal directional interval is defined as the directional interval extending to both sides of the direction perpendicular to the organic fertilizer strip application direction within a preset angular tolerance range. The orthogonal directional interval includes the direction perpendicular to the organic fertilizer strip application direction and all gradient directions whose deviation from the direction perpendicular to the organic fertilizer strip application direction does not exceed the preset angular tolerance range. The preset angle tolerance range is set based on the directional concentration characteristics of micro-topography texture. The setting is based on the following: the surface micro-topography undulations formed by strip application of organic fertilizer extend along the direction of organic fertilizer strip application and alternate along the direction orthogonal to the direction of organic fertilizer strip application. The pixel brightness gradient direction caused by micro-topography is highly concentrated in the orthogonal direction interval, while the gradient direction distribution caused by the difference in actual pasture biomass is more dispersed. The value of the preset angle tolerance range should be sufficient to accommodate the width of the gradient direction distribution caused by micro-topography, while avoiding excessive inclusion of randomly distributed gradient directions. For example, if the preset angle tolerance range is set to 15 degrees, the parallel direction interval covers all gradient directions within ±15 degrees of the direction of organic fertilizer strip application, and the orthogonal direction interval covers all gradient directions within ±15 degrees of the direction perpendicular to the direction of organic fertilizer strip application.

[0026] The distribution histogram of gradient directions within the parallel direction interval is statistically analyzed, and the first concentration is calculated. The angular range of the parallel direction interval is divided into multiple angular sub-intervals with a preset angular step size. The preset angular step size is determined based on the accuracy and statistical stability of the gradient direction calculation; for example, a preset angular step size of 5 degrees results in a 5-degree width for each angular sub-interval. The number of pixels in the orthophoto that fall within each angular sub-interval of the parallel direction interval is counted, forming a distribution histogram of gradient directions within the parallel direction interval. The first concentration C1 is calculated as: C1 = N1 / Ntotal; where N1 represents the total number of pixels falling within the parallel direction interval in the histogram of gradient directions within the parallel direction interval, and Ntotal represents the total number of pixels in the orthophoto. The distribution histogram of gradient directions within the orthogonal direction interval is statistically analyzed, and the second concentration is calculated. The angular range of the orthogonal direction interval is divided into multiple angular sub-intervals with a preset angular step size, the same as the preset angular step size for the parallel direction interval; for example, a preset angular step size of 5 degrees results in a 5-degree width for each angular sub-interval. The number of pixels in the orthophoto where the gradient direction θ falls within each angular sub-interval of the orthogonal direction interval is counted, forming a histogram of gradient direction distribution within the orthogonal direction interval. The second concentration C2 is calculated as: C2 = N2 / Ntotal; where N2 represents the total number of pixels in the histogram of gradient direction distribution within the orthogonal direction interval that fall within the orthogonal direction interval.

[0027] The absolute value of the difference between the first concentration C1 and the second concentration C2 is compared with a preset concentration difference threshold. When the absolute value of the difference between the first concentration C1 and the second concentration C2 exceeds the preset concentration difference threshold, it is determined that there is alternating light and dark texture in the orthophoto. The preset concentration difference threshold is set based on the random distribution characteristics of the gradient direction in the orthophoto under the condition of no alternating light and dark texture. The setting method is as follows: in the orthophoto without alternating light and dark texture, the gradient direction θ is approximately uniformly distributed in the range of 0 degrees to 180 degrees, the angular width of the parallel direction interval and the orthogonal direction interval are equal, and the difference in the number of pixels in the parallel direction interval and the number of pixels in the orthogonal direction interval is not significant. The statistical distribution of the absolute value of the difference between the first concentration and the second concentration is calculated on the orthophoto of a batch of pasture plots with known untreated organic fertilizer, and the 95th percentile of the statistical distribution is taken as the preset concentration difference threshold. When alternating light and dark texture is determined to exist, the process proceeds to S3 to further distinguish whether the cause of alternating light and dark texture is due to differences in micro-topographic reflection or differences in pasture biomass. Compared to the conventional method of extracting texture period by directly performing frequency domain transformation on orthophotos, this determination method utilizes the physical characteristic that the micro-topographic texture of strip-applied organic fertilizer has directional selectivity in gradient direction space, thus avoiding misjudgment caused by the overlap of pasture biomass difference texture and micro-topographic texture in spatial frequency due to frequency domain methods.

[0028] In the specific implementation of the embodiment corresponding to S3, the texture direction is determined from the region where the alternating light and dark texture is located in the orthophoto. The region where the alternating light and dark texture is located is the region in the orthophoto where the pixel gradient direction distribution shows directional clustering, as determined in S2. The range of the region where the alternating light and dark texture is located is determined by the spatial range covered by the pixels falling into the parallel direction interval in the histogram of gradient direction distribution within the parallel direction interval in S2. The gradient direction of each pixel in the region where the alternating light and dark texture is located is obtained. The gradient direction comes from the gradient direction θ of each pixel calculated using the Sobel operator in S2. The gradient direction dominance distribution interval is determined using the first concentration C1 and the second concentration C2 calculated in S2. The determination method is as follows: compare the size of the first concentration C1 and the second concentration C2. If the second concentration C2 is greater than the first concentration C1, then the orthogonal direction interval is taken as the gradient direction dominance distribution interval; if the first concentration C1 is greater than the second concentration C2, then the parallel direction interval is taken as the gradient direction dominance distribution interval.

[0029] The surface micro-topography undulations formed by strip application of organic fertilizer extend along the direction of the fertilizer application and alternate in directions orthogonal to the application direction. The pixel brightness gradient direction of the alternating light and dark texture caused by the micro-topography is highly concentrated within the orthogonal direction interval. Therefore, under the condition that alternating light and dark texture is determined to exist in S2, the second concentration C2 is greater than the first concentration C1, and the gradient direction dominance distribution interval is the orthogonal direction interval. The median gradient direction within the gradient direction dominance distribution interval is extracted as the main texture direction. The extraction method is as follows: statistically analyze the gradient direction distribution of each pixel in the area where the alternating light and dark texture is located within the gradient direction dominance distribution interval, and take the median of the direction distribution as the main texture direction. The main texture direction reflects the direction of the most drastic pixel brightness change in the alternating light and dark texture, and is used as the texture direction of the alternating light and dark texture.

[0030] Grayscale profiles are extracted line by line along the direction perpendicular to the texture direction, and the pixel grayscale sequence and pixel spatial coordinates of each profile are recorded. The texture direction is the main texture direction determined in the previous step, and the direction perpendicular to the texture direction is the direction after rotating the main texture direction by 90 degrees. A grayscale profile is a sequence of pixel grayscale values ​​arranged sequentially along a straight line perpendicular to the texture direction. The spacing between adjacent pixels in the grayscale profile is equal to the pixel size of the orthophoto. When extracting grayscale profiles, starting from one boundary of the region containing alternating light and dark textures, the process moves line by line along the texture direction with the pixel size of the orthophoto as the step size until the other boundary of the region containing alternating light and dark textures is covered. The start and end coordinates of each grayscale profile are restricted to the area containing the alternating light and dark textures. The pixel grayscale sequence of each grayscale profile is recorded as the pixel brightness value of each pixel arranged sequentially from the start to the end on the grayscale profile. The pixel spatial coordinates of each grayscale profile are recorded as the row number and column number of each pixel arranged sequentially from the start to the end on the grayscale profile in the orthophoto pixel coordinate system.

[0031] For each grayscale profile, the gradient sequence is obtained by calculating the first-order difference of the pixel grayscale sequence. The first-order difference is calculated as the difference in brightness value between two adjacent pixels on the grayscale profile. For a grayscale profile containing M pixels, the gradient sequence contains M-1 gradient values. The u-th gradient value D(u) is calculated as: D(u) = V(u+1) - V(u); where V(u+1) represents the brightness value of the (u+1)-th pixel on the grayscale profile, and V(u) represents the brightness value of the u-th pixel on the grayscale profile, with u ranging from 1 to M-1. The gradient sequence reflects the rate of change of pixel brightness values ​​on the grayscale profile along the direction perpendicular to the texture.

[0032] The pixel position corresponding to the gradient maxima in the gradient sequence is used as the position of the light-dark transition edge. The gradient maxima is the gradient value with the largest absolute value in the gradient sequence. When there are multiple gradient maxima with the same absolute value in the gradient sequence, the pixel position corresponding to the first occurrence of the gradient maxima is taken as the position of the light-dark transition edge. The pixel position corresponding to the gradient maxima is represented by the pixel index on the grayscale profile. The pixel index marks the position where the brightness transitions most sharply from dark to light or from light to dark in the alternating light-dark texture.

[0033] The edge transition area is defined by expanding outwards from the edge of the light-dark transition at a preset ratio. The preset ratio is set based on the spatial range characteristics of the light-dark transition caused by micro-topography. The setting is based on the relatively fixed ratio of the width of the shadow boundary transition area caused by micro-topography to the width of a single fertilization trench. The width of a single fertilization trench can be obtained by measuring the spatial period of the alternating light and dark texture in the orthophoto. The preset ratio is a fixed value between 10% and 30% of the width of a single fertilization trench, for example, 20% of the width of a single fertilization trench. The preset ratio is then expanded towards both the decreasing and increasing brightness sides of the light-dark transition edge. The decreasing brightness side is the direction in which the pixel brightness value decreases on the grayscale profile, and the increasing brightness side is the direction in which the pixel brightness value increases on the grayscale profile. The pixel range between the boundary of the decreasing brightness side and the boundary of the increasing brightness side after expansion constitutes the edge transition area.

[0034] The full width at half maximum (FWHM) of the gradient sequence within the edge transition region is calculated as the edge gradient width. The absolute value curve of the gradient sequence within the edge transition region is located at half the maximum absolute value of the gradient at half the height. The full width at half the height is the width of the region on the absolute value curve of the gradient sequence within the edge transition region that is greater than or equal to half the maximum absolute value of the gradient along the grayscale profile direction. The full width at half the height is expressed in pixels. The average edge gradient width of multiple grayscale profiles is compared with a preset width threshold. Multiple grayscale profiles cover all grayscale profiles along the texture direction within the region containing the alternating light and dark textures; the average is the arithmetic mean of the edge gradient widths of all grayscale profiles.

[0035] The preset width threshold is set based on the difference in edge sharpness between the light-dark transition edges caused by micro-topographic shadows and the gradual transition edges caused by differences in pasture biomass. The setting method is as follows: the light-dark transition edges caused by micro-topographic shadows are caused by the obstruction and reflection of direct light by the surface geometry, resulting in a narrow edge transition area and a small edge gradient width; the transition edges caused by differences in pasture biomass are caused by the gradual change in pasture coverage due to nutrient gradients, resulting in a wide edge transition area and a large edge gradient width. The statistical distribution of the edge gradient width is calculated on a batch of orthophotos that have been manually interpreted as having alternating light and dark textures caused by micro-topographic shadows after a known strip application of organic fertilizer. The upper limit of the statistical distribution is taken as the preset width threshold.

[0036] If the mean value is less than the preset width threshold, the alternating light and dark texture is determined to belong to the micro-topographic reflection difference and S4 is executed; otherwise, the orthophoto is directly used as the analysis image and S5 is executed. This determination method utilizes the physical difference in spatial transition gradient between the micro-topographic shadow edge and the pasture biomass gradient edge. Compared with the conventional method of directly performing image enhancement and threshold segmentation on the orthophoto, it can avoid false shadow edge misjudgment caused by local density changes in the pasture canopy.

[0037] In the specific implementation of the embodiment corresponding to S4, a brightness probability density distribution curve is established pixel by pixel for the pixels where the alternating light and dark textures belonging to micro-topographic reflection differences are determined. The pixels where the alternating light and dark textures belonging to micro-topographic reflection differences are determined are the set of all pixels in the area where the alternating light and dark textures are located after the alternating light and dark textures are determined to belong to micro-topographic reflection differences in S3. The method of establishing the brightness probability density distribution curve pixel by pixel is as follows: for each pixel in the pixels where the alternating light and dark textures are located, a spatial neighborhood window centered on that pixel is taken. The size of the spatial neighborhood window is determined according to the spatial scale of the micro-topographic undulations. The spatial neighborhood window needs to cover the range of the front slope and back slope within a single micro-topographic undulation unit. For example, the side length of the spatial neighborhood window is taken as 1.5 times the spatial period of the alternating light and dark textures in the orthophoto. The spatial period of the alternating light and dark textures is obtained by calculating the average distance between the pixel positions corresponding to the adjacent gradient maxima in the direction perpendicular to the texture direction of the grayscale profile in S3. The unit of the average distance is the number of pixels. The pixel brightness values ​​of all pixels within the spatial neighborhood window are statistically analyzed. A brightness probability density distribution curve is generated with pixel brightness values ​​on the horizontal axis and the frequency of occurrence corresponding to each pixel brightness value on the vertical axis. The horizontal axis of the brightness probability density distribution curve is in units of pixel brightness values, and the vertical axis is in units of frequency.

[0038] Calculate the skewness and kurtosis coefficients of the luminance probability density distribution curve. The formula for calculating the skewness coefficient S is: S = [n / ((n-1)(n-2))] × ∑[(Xi-Xavg) / σ] 3 Where n represents the total number of pixels in the spatial neighborhood window, Xi represents the pixel brightness value of the i-th pixel in the spatial neighborhood window, Xavg represents the arithmetic mean of the pixel brightness values ​​of all pixels in the spatial neighborhood window, σ represents the standard deviation of the pixel brightness values ​​of all pixels in the spatial neighborhood window, and ∑ represents the summation operation. The skewness coefficient S is dimensionless.

[0039] K is: K=[n(n+1) / ((n-1)(n-2)(n-3))]×∑[(Xi-Xavg) / σ] 4 -[3(n-1) 2 / ((n-2)(n-3))]; K represents the kurtosis coefficient; where n represents the total number of pixels in the spatial neighborhood window, Xi represents the luminance value of the i-th pixel in the spatial neighborhood window, Xavg represents the arithmetic mean of the luminance values ​​of all pixels in the spatial neighborhood window, σ represents the standard deviation of the luminance values ​​of all pixels in the spatial neighborhood window, and ∑ represents the summation operation. The unit of the kurtosis coefficient K is dimensionless. The kurtosis value of the normal distribution is 3, and the deviation of the kurtosis coefficient from the kurtosis value of the normal distribution is K-3.

[0040] The sign of the skewness coefficient distinguishes between pixels on the illuminated slope and pixels on the shady slope. When the skewness coefficient S is greater than 0, the peak of the luminance probability density distribution curve is biased towards the low-luminance side, and a long tail exists on the high-luminance side. Within the spatial neighborhood window, the pixels are mainly low-luminance shady slope pixels, with a few high-luminance illuminated slope pixels distributed in the long tail; this pixel is identified as a shady slope pixel. When the skewness coefficient S is less than 0, the peak of the luminance probability density distribution curve is biased towards the high-luminance side, and a long tail exists on the low-luminance side. Within the spatial neighborhood window, the pixels are mainly high-luminance illuminated slope pixels, with a few low-luminance shady slope pixels distributed in the long tail; this pixel is identified as an illuminated slope pixel. The absolute value of the skewness coefficient quantifies the degree of scattering asymmetry. The absolute value of the skewness coefficient indicates the degree to which the luminance probability density distribution curve deviates from a symmetrical distribution. The larger the absolute value of the skewness coefficient, the stronger the scattering asymmetry, and the greater the luminance difference between illuminated slope pixels and shady slope pixels.

[0041] The kurtosis coefficient's deviation from the kurtosis value of the normal distribution characterizes the concentration of scattering patterns. When the kurtosis coefficient K is greater than 3, the deviation K⁻³ from the normal distribution is positive, the luminance probability density distribution curve is steeper than the normal distribution, the pixel luminance values ​​within the spatial neighborhood window are highly concentrated near the arithmetic mean, and the scattering pattern is dominated by single scattering, with concentrated reflection in a specific direction. When the kurtosis coefficient K is less than 3, the deviation K⁻³ from the normal distribution is negative, the luminance probability density distribution curve is flatter than the normal distribution, the pixel luminance values ​​within the spatial neighborhood window are dispersed, and the scattering pattern is dominated by multiple scattering, with multiple scattering causing light to be evenly distributed in multiple directions. When the kurtosis coefficient K equals 3, the deviation K⁻³ from the normal distribution is 0, the pixel luminance values ​​within the spatial neighborhood window follow a normal distribution, and the scattering pattern does not show obvious single or multiple scattering dominance.

[0042] The weights of anisotropic scattering components are constructed by fusing the degree of scattering asymmetry and the concentration of scattering modes. The formula for calculating the weight W of the anisotropic scattering component is: W = α × |S| + β × max(K-3,0); where |S| represents the absolute value of the skewness coefficient, max(K-3,0) represents the deviation of the kurtosis coefficient from the kurtosis value of the normal distribution, K-3 is taken as the maximum value of 0, α represents the contribution factor of the degree of scattering asymmetry, and β represents the contribution factor of the concentration of scattering modes. The scattering asymmetry contribution factor α and the scattering mode concentration contribution factor β are set according to the intensity requirements of micro-topographic reflection difference correction. The setting method is as follows: the scattering asymmetry reflects the brightness contrast between the sunlit slope and the shaded slope, and the scattering mode concentration reflects the dominance of a single scattering. The two together determine the contribution ratio of micro-topographic reflection difference to the brightness value of each pixel. The values ​​of the scattering asymmetry contribution factor α and the scattering mode concentration contribution factor β need to constrain the weight W of the anisotropic scattering component between 0 and 1. For example, α is 0.6 and β is 0.4. When both |S| and max(K-3,0) are large, the weight W of the anisotropic scattering component is close to 1, indicating that the contribution of micro-topographic reflection difference is large. When both |S| and max(K-3,0) are small, the weight W of the anisotropic scattering component is close to 0, indicating that the contribution of micro-topographic reflection difference is small. The physical basis of the max(K-3,0) operation is: when the kurtosis coefficient K is less than or equal to 3, the scattering mode is dominated by multiple scattering or normal distribution, there is no anisotropic reflection difference caused by single scattering concentration, and the contribution of the kurtosis coefficient to the micro-topographic reflection difference is 0.

[0043] Anisotropic scattering component weighting is used to remove the anisotropic scattering component and retain the Lambertian reflection component from the original pixel brightness value. The original pixel brightness value is the pixel brightness value in the orthophoto. Anisotropic scattering component removal is calculated by multiplying the original pixel brightness value by the anisotropic scattering component weight to obtain the anisotropic scattering component. Lambertian reflection component retention is calculated by subtracting the anisotropic scattering component from the original pixel brightness value to obtain the Lambertian reflection component. The Lambertian reflection component is the brightness value of the pixel under uniformly scattered illumination conditions after eliminating micro-topographic reflection differences.

[0044] The Lambert reflectance components, after being merged and stripped, yield a corrected image that eliminates micro-topographical reflectance differences. This corrected image is then used as the analysis image. The merging method involves filling the calculated Lambert reflectance components for each pixel containing alternating light and dark textures into a corrected image matrix of the same size as the orthorectified image, according to the pixel's spatial coordinates. The brightness values ​​of pixels not containing alternating light and dark textures in the corrected image matrix are directly copied from the corresponding pixel brightness values ​​in the orthorectified image. After merging, a corrected image eliminating micro-topographical reflectance differences is obtained. In this corrected image, the anisotropic scattering components of pixels containing alternating light and dark textures are stripped, while the Lambert reflectance components are retained, thus eliminating the visual differences in light and dark caused by micro-topography. This corrected image is then used as the analysis image and transferred to S5 for vegetation index extraction. Compared to conventional methods such as directly performing histogram equalization on orthophotos or using bandpass filtering based on fixed kernels, this reflectivity component separation method adaptively extracts micro-topographic scattering features from the statistical distribution of brightness using skewness and kurtosis coefficients. This avoids the problems of insufficient or excessive correction under different terrain undulations caused by fixed parameter methods. Furthermore, the separated anisotropic scattering components have clear physical meanings, corresponding to the physical process of enhanced single-scattering on the sunlit slope and weakened multiple-scattering on the shaded slope.

[0045] In the specific implementation of the embodiment corresponding to S5, the spectral reflectance of the pasture canopy cover pixels in the analysis image is obtained. The analysis image is the corrected image obtained in S4 after eliminating micro-topographical reflectance differences. The spectral reflectance of each pixel in the analysis image includes spectral reflectance values ​​of multiple bands, including at least the red edge band, near-infrared band, green band, and red band. The wavelength range of the red edge band is 700 nm to 750 nm, the wavelength range of the near-infrared band is 750 nm to 900 nm, the wavelength range of the green band is 520 nm to 590 nm, and the wavelength range of the red band is 620 nm to 700 nm. The pasture canopy cover pixels are the pixels within the spatial range of pasture canopy cover in the analysis image. The spatial range of pasture canopy cover is extracted from the analysis image using a vegetation segmentation method. The vegetation segmentation method uses the Otsu method to calculate the segmentation threshold of the normalized vegetation index (NDI) of the analysis image. Pixels with a NDI greater than the segmentation threshold are identified as pasture canopy cover pixels.

[0046] The Normalized Difference Vegetation Index (NDVI) is obtained by calculating the ratio of red-edge and near-infrared bands of spectral reflectance. The formula for NDVI is: NDVI = (RNIR - RRedEdge) / (RNIR + RRedEdge); where RNIR represents the spectral reflectance value in the near-infrared band, and RRedEdge represents the spectral reflectance value in the red-edge band. The NDVI value ranges from -1 to 1, reflecting the combined state of chlorophyll content and canopy coverage of pasture canopies. The Visible Light Atmospheric Resistance Index (VARI) is obtained by calculating the normalized difference between green and red bands of spectral reflectance. The formula for VARI is: VARI = (RGreen - RRed) / (RGreen + RRed); where RGreen represents the spectral reflectance value in the green band, and RRed represents the spectral reflectance value in the red band. The Visible Light Resistance Index (VARI) ranges from -1 to 1. The VARI reflects the proportion of green vegetation cover in the pasture canopy and has the ability to resist the influence of atmospheric scattering.

[0047] When generating a vegetation index reflecting the growth status of pasture canopy by multiplying and combining the normalized vegetation index and the visible light atmospheric resistance index, the corrected pixel regions identified as belonging to micro-topographic reflectance differences and the non-differential pixel regions not identified as belonging to micro-topographic reflectance differences in the analyzed image are obtained. The corrected pixel region is the spatial coverage of the pixels in the analyzed image containing the alternating light and dark textures identified as belonging to micro-topographic reflectance differences in S4. The corresponding anisotropic scattering component weight of each pixel in the corrected pixel region is calculated in S4. The non-differential pixel region is the spatial range of pasture canopy-covered pixels in the analyzed image other than the corrected pixel regions.

[0048] The first product of the normalized vegetation index (NDI) and the visible light atmospheric resistance index (VRI) of the corrected pixel region is calculated using the formula: P1 = NDVIcorr × VARIcorr; where NDVIcorr represents the NDI of the pixel in the corrected pixel region, and VARIcorr represents the VRI of the pixel in the corrected pixel region. The second product of the NDI and the VRI of the non-differential pixel region is calculated using the formula: P2 = NDVIuncorr × VARIuncorr; where NDVIuncorr represents the NDI of the pixel in the non-differential pixel region, and VARIuncorr represents the VRI of the pixel in the non-differential pixel region.

[0049] The first product is assigned a correction weight coefficient determined by the weights of the anisotropic scattering components in S4. The correction weight coefficient ω is determined by substituting the weights W of the anisotropic scattering components corresponding to the pixels in S4 into the weight transformation formula ω = 1 - W. The value of the correction weight coefficient ω ranges from 0 to 1. The weights W of the anisotropic scattering components in S4 reflect the proportion of contribution of micro-topographic reflection differences to the pixel brightness value. A larger weight W indicates a greater contribution from micro-topographic reflection differences, resulting in more anisotropic scattering components being stripped from the pixel in S4. This leads to a relative weakening of the pasture canopy signal in the remaining Lambertian reflection components, and the correction weight coefficient ω is correspondingly reduced to compensate for the signal attenuation effect caused by stripping anisotropic scattering components in vegetation index extraction. The second product is assigned a unit weight coefficient, with a value of 1. A unit weight coefficient indicates that the vegetation index product of non-differential pixel areas uses the original calculation results without weighted adjustment.

[0050] The weighted first product and the weighted second product are combined to generate a vegetation index reflecting the growth status of the pasture canopy. The specific formula for calculating the vegetation index VI reflecting the growth status of the pasture canopy is: VI = ω × P1 + 1 × P2; where ω represents the correction weight coefficient, P1 represents the first product, and P2 represents the second product. The combined vegetation index VI reflecting the growth status of the pasture canopy uses the weighted first product in the corrected pixel area and the original second product in the non-differential pixel area. The vegetation index VI reflecting the growth status of the pasture canopy is spatially continuous throughout the entire pasture plot. Compared to conventional methods that directly calculate the normalized vegetation index or visible light atmospheric resistance index uniformly for the entire analysis image, this product combination generation method introduces anisotropic scattering component weights as adjustment factors for vegetation index calculation in the correction pixel area. This solves the problem of systematically low vegetation index in the correction pixel area due to the stripping of anisotropic scattering components after S4 reflectance component separation, ensuring that the vegetation index of the correction pixel area and the non-differential pixel area are spatially comparable.

[0051] In the specific implementation of the embodiment corresponding to S6, the spatial distribution array of vegetation indices reflecting the growth status of the pasture canopy generated in S5 is obtained. The spatial distribution array of vegetation indices reflecting the growth status of the pasture canopy is a two-dimensional numerical matrix. The number of rows and columns of the two-dimensional numerical matrix are equal to the number of pixel rows and columns of the orthophoto of the pasture patch generated in S1, respectively. Each element in the two-dimensional numerical matrix stores the value of the vegetation index reflecting the growth status of the pasture canopy at the corresponding spatial location. The value of the vegetation index reflecting the growth status of the pasture canopy comes from the vegetation index reflecting the growth status of the pasture canopy generated in S5 by merging the weighted first product of the corrected pixel region and the original second product of the non-differential pixel region.

[0052] A spatial reference frame for the vegetation spatial distribution map is constructed based on the pixel grid of the orthophoto of pasture plots. The orthophoto of pasture plots is obtained from the orthophoto of pasture plots acquired in S1. The pixel grid of the orthophoto of pasture plots consists of the arrangement of pixels along the row and column directions of the orthophoto. The spatial reference frame for the vegetation spatial distribution map uses the number of rows and columns of the pixel grid in the orthophoto of pasture plots as the number of rows and columns of the spatial reference frame. The position of each pixel in the spatial reference frame of the vegetation spatial distribution map corresponds to the same spatial position as the pixel in the same row and column of the orthophoto of pasture plots. The spatial reference frame for the vegetation spatial distribution map ensures that the generated vegetation spatial distribution map of pasture plots is strictly consistent with the orthophoto of pasture plots in spatial location.

[0053] Each vegetation index value in the spatial distribution array of vegetation indices is mapped pixel-by-pixel to the corresponding pixel grid position in the spatial reference frame. The mapping method is as follows: for the element in the p-th row and q-th column of the spatial distribution array of vegetation indices, the stored vegetation index value reflecting the growth status of the pasture canopy is assigned to the pixel grid position in the p-th row and q-th column of the spatial reference frame of the vegetation growth spatial distribution map. The value of p ranges from 1 to the number of pixel rows in the orthophoto of the pasture patch, and the value of q ranges from 1 to the number of pixel columns in the orthophoto of the pasture patch. After mapping, each pixel grid position in the spatial reference frame of the vegetation growth spatial distribution map carries the vegetation index value reflecting the growth status of the pasture canopy at that spatial location.

[0054] The mapped pixel grid is then rendered using partitioned, graded, and color-coded rendering to generate a spatial distribution map of pasture growth. The partitioned, graded, and color-coded rendering method involves dividing the numerical range of the vegetation index reflecting the pasture canopy growth status into multiple numerical intervals, each corresponding to a different level of pasture growth. The division of the vegetation index's numerical range is based on the degree of difference in pasture canopy growth status. The division method involves evenly dividing the vegetation index's numerical range from minimum to maximum value into a preset number of numerical intervals. The preset number is set according to the required level of detail in pasture growth monitoring; for example, a preset number of 5 intervals corresponds to five levels of pasture growth: poor, relatively poor, moderate, good, and excellent. The boundary values ​​between adjacent numerical intervals are the minimum value plus the value of the fifth division point of the numerical range span.

[0055] A rendering color is assigned to each numerical range, following a color gradient from low to high levels. For example, low-level pasture growth corresponds to red, medium-level pasture growth to yellow, and high-level pasture growth to green. For each pixel in the mapped grid, the numerical range in which the vegetation index value reflecting the pasture canopy growth status falls is determined, and that pixel is rendered with the color corresponding to that range. After rendering all pixel locations, a spatial distribution map of pasture growth is generated. This map visually presents the differences in pasture canopy growth status at various spatial locations within the pasture plot using color, eliminating the interference of micro-topographical reflection differences caused by strip application of organic fertilizer on growth interpretation.

[0056] Example 2: Figure 2 A schematic diagram of an image recognition-based organic fertilizer return-to-field forage growth monitoring system of the present invention is provided. The system includes: The image acquisition module is used to acquire orthophotos of pasture plots and simultaneously record the direction of organic fertilizer strip application; The texture detection module is used to calculate the first concentration of the pixel gradient direction in the orthophoto image within the direction interval where the organic fertilizer strip application direction is located, and the second concentration within the direction interval orthogonal to the organic fertilizer strip application direction. When the difference between the first concentration and the second concentration exceeds the preset concentration difference threshold, it is determined that there is a texture with alternating light and dark and the cause discrimination module is executed. Otherwise, the orthophoto image is used as the analysis image and the index extraction module is executed. The cause discrimination module is used to extract grayscale profiles along the direction of the vertical alternating light and dark texture, calculate the edge gradient width of the grayscale profile, and if the edge gradient width is less than the preset width threshold, it is determined that the alternating light and dark texture belongs to micro-topographic reflection difference and the reflection decoupling module is executed; otherwise, the orthophoto is used as the analysis image and the exponential extraction module is executed. The reflection decoupling module is used to establish the brightness probability density distribution of the pixels containing alternating light and dark textures. Based on the skewness coefficient and kurtosis coefficient of the brightness probability density distribution, the reflectivity components of each pixel are separated to obtain a corrected image that eliminates micro-topographic reflection differences as the analysis image. The index extraction module is used to extract vegetation indices reflecting the growth status of pasture canopies from the analyzed images; The distribution construction module is used to construct a spatial distribution map of the growth of pasture plots based on vegetation indices.

[0057] All calculations involved in the embodiments are performed using dimensionless numerical values, and the preset parameters and thresholds in the calculations can be set by those skilled in the art according to actual conditions.

[0058] This technical solution can be flexibly deployed, for example, as embedded software running on device hardware, or installed on personal computers or other smart terminals with user interfaces, thus adapting to various hardware environments and usage requirements.

[0059] The above solutions can be implemented in software, hardware, firmware, or a combination thereof. When implemented in software, they are presented, in whole or in part, as a computer program product, containing one or more computer instructions or programs. When these instructions or programs are loaded and executed on a computer, results are produced corresponding to the processes or functions of the embodiments of this application. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. Computer instructions can be stored in a computer-readable storage medium or transmitted from one medium to another via wired or wireless means, such as from a website, server, or data center via wired means such as fiber optic cables, twisted-pair cables, or coaxial cables, or wireless means such as infrared or microwaves to another site. A computer-readable storage medium refers to any usable medium that a computer can access or a data storage device such as a server or data center containing one or more usable media, including magnetic media such as floppy disks, hard disks, and magnetic tapes, optical media such as DVDs, and semiconductor media such as solid-state drives.

[0060] The specific working process of the system, device and module can be found in the method embodiment, and will not be repeated here.

[0061] The disclosed systems, devices, and methods can be implemented in other ways. The device embodiments are for illustrative purposes only, and the module division is only a logical division. In practice, different divisions can be implemented, such as merging or integrating multiple modules or components, or omitting some features. Coupling, direct coupling, or communication connections between the components can be achieved through interfaces, while indirect coupling or communication connections can take electrical, mechanical, or other forms.

[0062] The modules described as separate components may or may not be physically separated. The components shown as modules can be physical hardware or software, and can be deployed centrally or distributed across multiple network nodes. Some or all of the modules can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0063] The functional modules in each embodiment can be integrated into one processing module, or they can exist independently, or two or more modules can be integrated into one.

[0064] If the functionality is implemented as a software module and used as an independent product, it can be stored in a computer-readable storage medium. Under this understanding, the substantial contribution of the technical solution of this application can be embodied in the form of a software product. This computer software product is stored in a storage medium and contains instructions to cause a computer device, such as a personal computer, server, or network device, to execute all or part of the steps of the methods in the embodiments of this application. The storage medium includes any medium capable of storing program code, such as a USB flash drive, portable hard drive, read-only memory, random access memory, magnetic disk, or optical disk.

[0065] The above are merely specific embodiments of this application, and the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should fall within the scope of protection of this application.

Claims

1. An image recognition-based organic fertilizer incorporation pasture grass vigor monitoring method, characterized in that, include: S1: Acquire orthophotos of pasture plots and simultaneously record the direction of organic fertilizer strip application; S2: Calculate the first concentration of the pixel gradient direction in the orthophoto image within the direction interval where the organic fertilizer strip application direction is located, and the second concentration within the direction interval orthogonal to the organic fertilizer strip application direction. When the difference between the first concentration and the second concentration exceeds the preset concentration difference threshold, it is determined that there is a light and dark alternating texture and S3 is executed; otherwise, the orthophoto image is used as the analysis image and S5 is executed. S3: Extract grayscale profiles along the direction of the alternating light and dark texture, calculate the edge gradient width of the grayscale profiles, and if the edge gradient width is less than the preset width threshold, determine that the alternating light and dark texture belongs to micro-topographical reflection difference and execute S4; otherwise, use the orthophoto as the analysis image and execute S5. S4: Establish the luminance probability density distribution for the pixels containing alternating light and dark textures, and separate the reflectance components of each pixel based on the skewness coefficient and kurtosis coefficient of the luminance probability density distribution to obtain a corrected image that eliminates micro-topographic reflection differences as the analysis image. S5: Extract vegetation indices reflecting the growth status of pasture canopy from analyzed images; S6: Construct a spatial distribution map of the growth of pasture plots using vegetation indices.

2. The method for monitoring the growth of forage grass after organic fertilizer is returned to the field based on image recognition, as described in claim 1, is characterized in that, S1 includes: Under cloudy and uniformly diffused light conditions, multi-view image sequences are acquired from above the pasture plots at a vertically downward shooting angle and a preset spatial resolution. Motion recovery structure processing is performed on the acquired multi-view image sequences to generate a three-dimensional point cloud of the pasture plots. Orthorectification is performed on the three-dimensional point cloud of the pasture plots to obtain an orthophoto of the pasture plots. The direction of organic fertilizer strip application is recorded while acquiring the orthophoto of the pasture plots.

3. The method for monitoring the growth of forage grass after organic fertilizer is returned to the field based on image recognition according to claim 1, characterized in that, S2 includes: using the Sobel operator to calculate the gradient magnitude and gradient direction of each pixel in the orthophoto; dividing the direction interval with the direction of organic fertilizer strip application as the reference direction into parallel direction intervals and orthogonal direction intervals; statistically analyzing the distribution histogram of gradient directions in the parallel direction interval and calculating the first concentration; statistically analyzing the distribution histogram of gradient directions in the orthogonal direction interval and calculating the second concentration; when the difference between the first concentration and the second concentration exceeds a preset concentration difference threshold, it is determined that there is an alternating light and dark texture in the orthophoto.

4. The method for monitoring the growth of forage grass after organic fertilizer is returned to the field based on image recognition according to claim 1, characterized in that, S3 includes: determining the texture direction from the area where the alternating light and dark textures are located in the orthophoto; extracting grayscale profiles one by one along the direction perpendicular to the texture direction and recording the pixel grayscale sequence and pixel spatial coordinates of each grayscale profile; calculating the first-order difference of the pixel grayscale sequence of each grayscale profile to obtain the gradient sequence; locating the pixel position corresponding to the maximum gradient value in the gradient sequence as the position of the light-dark transition edge; expanding outwards from the position of the light-dark transition edge to both sides at a preset ratio to determine the edge transition area; calculating the full width at half maximum (FWHM) of the gradient sequence in the edge transition area as the edge gradient width; comparing the average edge gradient width of multiple grayscale profiles with a preset width threshold; if the average value is less than the preset width threshold, determining that the alternating light and dark texture belongs to micro-topographic reflection difference and executing S4; otherwise, directly using the orthophoto as the analysis image and executing S5.

5. The method for monitoring the growth of forage grass after organic fertilizer is returned to the field based on image recognition, as described in claim 4, is characterized in that... Determining the texture direction includes: obtaining the gradient direction of each pixel in the region where the alternating light and dark texture is located; using the first concentration and the second concentration calculated in S2 to determine the gradient direction dominance distribution range; extracting the median gradient direction within the gradient direction dominance distribution range as the main texture direction; and using the main texture direction as the texture direction of the alternating light and dark texture.

6. The method for monitoring the growth of forage grass after organic fertilizer is returned to the field based on image recognition according to claim 1, characterized in that, S4 includes: establishing a brightness probability density distribution curve for each pixel containing alternating light and dark textures that are determined to belong to micro-topographic reflection differences; calculating the skewness coefficient and kurtosis coefficient of the brightness probability density distribution curve; distinguishing between light-facing and shadow-facing pixels by the sign of the skewness coefficient; quantifying the degree of scattering asymmetry by the absolute value of the skewness coefficient; characterizing the concentration of scattering patterns by the deviation of the kurtosis coefficient from the kurtosis value of the normal distribution; constructing anisotropic scattering component weights by fusing the degree of scattering asymmetry and the concentration of scattering patterns; using the anisotropic scattering component weights to strip the anisotropic scattering components and retain the Lambertian reflection components from the original brightness values ​​of the pixels; merging the stripped Lambertian reflection components to obtain a corrected image that eliminates micro-topographic reflection differences; and using the corrected image as the analysis image.

7. The method for monitoring the growth of forage grass after organic fertilizer is returned to the field based on image recognition according to claim 1, characterized in that, S5 includes: acquiring the spectral reflectance of the pasture canopy cover pixels in the analysis image; calculating the ratio of the red edge band and the near-infrared band of the spectral reflectance to obtain the normalized vegetation index; calculating the normalized difference of the spectral reflectance between the green band and the red band to obtain the visible light atmospheric resistance index; and multiplying the normalized vegetation index and the visible light atmospheric resistance index to generate a vegetation index reflecting the growth status of the pasture canopy.

8. The method for monitoring the growth of forage grass after organic fertilizer is returned to the field based on image recognition, as described in claim 7, is characterized in that... The process of generating a vegetation index reflecting the growth status of pasture canopy includes: acquiring corrected pixel regions identified as belonging to micro-topographic reflectance differences and non-differential pixel regions not identified as belonging to micro-topographic reflectance differences in the analyzed image; calculating the first product of the normalized vegetation index and the visible light resistance index of the corrected pixel regions; calculating the second product of the normalized vegetation index and the visible light resistance index of the non-differential pixel regions; assigning a correction weight coefficient to the first product determined by the weight of the anisotropic scattering component in S4; assigning a unit weight coefficient to the second product; and combining the weighted first product and the weighted second product to generate a vegetation index reflecting the growth status of pasture canopy.

9. A method for monitoring the growth of forage grass after organic fertilizer is returned to the field based on image recognition, as described in claim 1, characterized in that, S6 include: A spatial distribution array of vegetation indices is obtained. A spatial reference frame for the growth spatial distribution map is constructed based on the pixel grid of the orthophoto of the pasture plots. Each vegetation index value in the spatial distribution array of vegetation indices is mapped pixel by pixel to the corresponding pixel grid position in the spatial reference frame. The mapped pixel grids are then partitioned, graded, and color-coded for rendering to generate a growth spatial distribution map of the pasture plots.

10. An image recognition-based system for monitoring the growth of forage grass after organic fertilizer application, used to implement the image recognition-based method for monitoring the growth of forage grass after organic fertilizer application as described in any one of claims 1-9, characterized in that, include: The image acquisition module is used to acquire orthophotos of pasture plots and simultaneously record the direction of organic fertilizer strip application; The texture detection module is used to calculate the first concentration of the pixel gradient direction in the orthophoto image within the direction interval where the organic fertilizer strip application direction is located, and the second concentration within the direction interval orthogonal to the organic fertilizer strip application direction. When the difference between the first concentration and the second concentration exceeds the preset concentration difference threshold, it is determined that there is a texture with alternating light and dark and the cause discrimination module is executed. Otherwise, the orthophoto image is used as the analysis image and the index extraction module is executed. The cause discrimination module is used to extract grayscale profiles along the direction of the vertical alternating light and dark texture, calculate the edge gradient width of the grayscale profile, and if the edge gradient width is less than the preset width threshold, it is determined that the alternating light and dark texture belongs to micro-topographic reflection difference and the reflection decoupling module is executed; otherwise, the orthophoto is used as the analysis image and the exponential extraction module is executed. The reflection decoupling module is used to establish the brightness probability density distribution of the pixels containing alternating light and dark textures. Based on the skewness coefficient and kurtosis coefficient of the brightness probability density distribution, the reflectivity components of each pixel are separated to obtain a corrected image that eliminates micro-topographic reflection differences as the analysis image. The index extraction module is used to extract vegetation indices reflecting the growth status of pasture canopies from the analyzed images; The distribution construction module is used to construct a spatial distribution map of the growth of pasture plots based on vegetation indices.

Images