A method and system for measuring vegetation cover

Abstract

The application discloses a kind of vegetation coverage measurement method and system, belong to vegetation coverage measurement technical field, comprising: for the heterogeneous image obtained by field handheld device shooting, using the image registration method based on feature matching, by extracting the significant feature points in image, and the similarity between feature points is calculated, the corresponding relationship between image is determined, and then the geometric correction and spatial registration of image are realized, eliminate the influence of angle difference and inconsistent resolution, obtain spatial consistent overall image;To ensure the accuracy of overall image, in the image fusion process, according to the feature point distribution of image overlapping area, the fusion weight is adaptively adjusted, the weighted average of pixel gray value in overlapping area is carried out, to ensure that the image after fusion is smoothly transitioned in overlapping area, avoid obvious splicing trace, improve the visual quality and measurement accuracy of overall image.

Description

Technical Field

[0001] This invention belongs to the field of vegetation coverage measurement technology, and particularly relates to a method and system for measuring vegetation coverage. Background Technology

[0002] Numerous technical challenges arise when using handheld field devices to measure vegetation cover. First, images captured by these devices often exhibit differences in perspective and resolution, making the effective fusion of these heterogeneous images to construct a complete and high-quality overall image dataset a significant challenge. Second, when extracting green vegetation, relying solely on color information is susceptible to interference from factors such as lighting and shadows, leading to inaccurate results. Meanwhile, recognition methods based on texture and shape features may suffer from high computational complexity and insufficient robustness.

[0003] Furthermore, even with the complete image obtained, verifying its accuracy remains a challenging issue. Traditional manual sampling verification methods are time-consuming and labor-intensive, while automated verification algorithms struggle to balance accuracy and efficiency. Finally, balancing local and global optimization during the comprehensive analysis and removal of overlapping portions to avoid over-segmentation or under-segmentation is also a problem worthy of in-depth exploration. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention proposes a method and system for measuring vegetation coverage, thereby resolving the issues present in the prior art.

[0005] To achieve the above objectives, the present invention provides a method for measuring vegetation coverage, comprising:

[0006] A depth camera is installed on a handheld device in the field, and a heterogeneous depth image of the area to be tested is acquired based on the depth camera.

[0007] The deep heterogeneous images are preprocessed to obtain spatially consistent overall images, and the spatially consistent overall images are fused to obtain a fused dataset.

[0008] A multi-feature fusion model for vegetation identification is constructed based on a random forest classifier. The fused dataset is input into the multi-feature fusion model for vegetation identification for calculation to obtain the binary segmentation result of the vegetation.

[0009] The region merging method based on graph theory and the minimum spanning tree algorithm are used to process the binary segmentation results of the vegetation to obtain the merged vegetation region.

[0010] Calculate the vegetation index value for each pixel in the merged vegetation region, generate a fine distribution map of vegetation based on the vegetation index value, and simultaneously calculate the vegetation coverage estimate of the merged vegetation region.

[0011] The detailed distribution map of the vegetation and the estimated vegetation coverage are combined to generate the final vegetation coverage measurement result.

[0012] Preferably, the process of preprocessing the depth heterogeneous image to obtain a spatially consistent overall image includes:

[0013] Significant feature points are extracted from the deep heterogeneous image, the similarity between the significant feature points is calculated, and geometric correction and spatial registration are performed on the deep heterogeneous image based on the similarity to obtain a spatially consistent overall image.

[0014] Preferably, the process of fusing the spatially consistent overall images to obtain a fused dataset includes:

[0015] Extract feature points from the spatially consistent overall image and calculate the distribution density of the feature points;

[0016] The overlapping region of the image is determined based on the distribution density of the feature points, and the initial fusion weight of each pixel in the overlapping region is calculated.

[0017] For each pixel in the overlapping area, obtain its corresponding gray value in the image to be fused, and perform a weighted average of the gray values ​​according to the initial fusion weights to obtain the fused gray value of the pixel.

[0018] The fused pixel gray values ​​in the overlapping area are then combined with the original pixel gray values ​​in the non-overlapping area to obtain a preliminary fused image.

[0019] The preliminary fused images are subjected to image enhancement processing to generate the fused dataset.

[0020] Preferably, the process of obtaining the binary segmentation result of vegetation includes:

[0021] After color conversion processing of the fused dataset, information extraction is performed to obtain color features, texture features, and shape features;

[0022] The color features, texture features, and shape features are normalized to obtain a multi-feature fusion vector;

[0023] A random forest classifier is constructed, and the random forest classifier is trained based on a historical dataset to obtain a multi-feature fusion model for vegetation recognition.

[0024] The multi-feature fusion vector is input into the multi-feature fusion model for vegetation recognition for calculation to obtain the binary segmentation result of the vegetation.

[0025] Preferably, the process of extracting information from the fused dataset after color conversion to obtain color features, texture features, and shape features includes:

[0026] After converting the fused dataset from RGB space to HSV space, the green channel information is extracted to obtain the color features of green vegetation;

[0027] The texture features of the fused dataset are calculated based on the gray-level co-occurrence matrix method to obtain the texture features of the vegetation area;

[0028] Based on the shape characteristics of the vegetation area, extract the shape features of the vegetation.

[0029] Preferably, the process of obtaining the merged vegetation area includes:

[0030] The graph theory-based region merging method transforms the binary segmentation result of the vegetation into an undirected weighted graph. Through the minimum spanning tree algorithm, regions with high similarity are gradually merged until all regions are merged into a whole and then output, generating the merged vegetation region.

[0031] Preferably, the process of generating a detailed vegetation distribution map based on the vegetation index value, and simultaneously calculating the estimated vegetation cover of the merged vegetation area, includes:

[0032] The vegetation area of ​​the merged vegetation region is obtained based on pixel statistics, the total proportion of the merged vegetation region is calculated, and a preliminary vegetation coverage estimate is obtained based on the total proportion.

[0033] Obtain the multispectral information of the overall image, and calculate the vegetation index value of each pixel based on the spectral information;

[0034] The NDVI value of the vegetation index for each pixel is calculated, and a vegetation index distribution map is generated based on the NDVI value.

[0035] The vegetation index distribution map is binarized to obtain a detailed vegetation distribution map.

[0036] Preferably, the process of generating the final vegetation cover measurement result includes:

[0037] The number of pixels with a value of 1 in the detailed distribution map of the vegetation is counted, and then divided by the total number of pixels in the image to obtain the detailed vegetation coverage rate based on the vegetation index.

[0038] The final vegetation coverage measurement result is obtained by weighting the preliminary vegetation coverage estimate and the fine vegetation coverage based on the vegetation index.

[0039] To achieve the above objectives, the present invention also provides a system for measuring vegetation coverage, comprising:

[0040] The data acquisition module is used to install a depth camera on a handheld device in the field and acquire a heterogeneous depth image of the area to be tested based on the depth camera;

[0041] The data fusion module is used to preprocess the deep heterogeneous images to obtain spatially consistent overall images, and to fuse the spatially consistent overall images to obtain a fused dataset.

[0042] The model calculation module is used to construct a multi-feature fusion model for vegetation recognition based on a random forest classifier. The fusion dataset is input into the multi-feature fusion model for vegetation recognition for calculation to obtain the binary segmentation result of the vegetation.

[0043] The merging module is used to process the binary segmentation results of the vegetation based on graph theory-based region merging methods and minimum spanning tree algorithms to obtain merged vegetation regions.

[0044] The estimation calculation module is used to calculate the vegetation index value of each pixel in the merged vegetation area, generate a fine distribution map of vegetation based on the vegetation index value, and calculate the vegetation coverage estimate of the merged vegetation area.

[0045] The result generation module is used to combine the detailed distribution map of the vegetation with the vegetation coverage estimate to generate the final vegetation coverage measurement result.

[0046] Compared with the prior art, the present invention has the following advantages and technical effects:

[0047] This invention discloses a method for measuring vegetation cover based on imagery from handheld field devices. For heterogeneous images captured by handheld field devices, a feature-matching image registration method is employed to achieve spatially consistent overall images. During image fusion, the fusion weights of overlapping regions are adaptively adjusted to ensure a smooth transition. For vegetation extraction, a random forest classification model is constructed by comprehensively utilizing multiple features such as color, texture, and shape. Overlapping parts are removed using a graph theory-based region merging method, and the final coverage is calculated by combining area ratio and vegetation index. To verify accuracy, high-precision satellite imagery is used as a reference for pixel-level comparison and confusion matrix analysis. This invention solves the problems of heterogeneous imagery from handheld field devices and low vegetation extraction accuracy, achieving high-precision measurement of vegetation cover. Through multi-source data fusion, multi-feature classification, and accuracy verification, the reliability and practicality of the measurement results are improved, providing an effective tool for vegetation monitoring and ecological assessment. Attached Figure Description

[0048] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings:

[0049] Figure 1 This is a flowchart of the vegetation coverage measurement method according to an embodiment of the present invention. Detailed Implementation

[0050] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0051] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0052] Example 1

[0053] like Figure 1 As shown, this embodiment provides a method for measuring vegetation coverage, including:

[0054] Step S101: For heterogeneous images captured by handheld devices in the field, an image registration method based on feature matching is adopted. By extracting significant feature points in the images and calculating the similarity between feature points, the correspondence between images is determined, thereby realizing geometric correction and spatial registration of the images, eliminating the influence of perspective differences and resolution inconsistencies, and obtaining a spatially consistent overall image.

[0055] In step S102, to ensure the accuracy of the overall image, during the image fusion process, the fusion weight is adaptively adjusted according to the distribution of feature points in the overlapping area of ​​the image. The pixel gray values ​​in the overlapping area are weighted and averaged to ensure that the fused image transitions smoothly in the overlapping area, avoids obvious stitching marks, and improves the visual quality and measurement accuracy of the overall image.

[0056] Step S103: When extracting green vegetation information, a multi-feature fusion model for vegetation recognition is constructed by comprehensively utilizing multiple features such as color, texture, and shape. Green channel information is extracted through color space transformation, texture features of the image are calculated, and shape features of the vegetation are extracted through morphological operations. These features are then input into a random forest classifier for training and prediction to obtain the binary segmentation result of the vegetation.

[0057] Step S104: To remove overlapping parts in the vegetation extraction results, a graph-based region merging method is used. The vegetation segmentation results are converted into an undirected weighted graph. The minimum spanning tree algorithm is used to gradually merge regions with high similarity until all regions are merged into a single entity. During the merging process, a similarity threshold is set to control the granularity of the merging, avoiding both over-segmentation and under-segmentation.

[0058] Step S105: For the merged vegetation area, calculate its area as a percentage of the entire image to obtain a preliminary vegetation cover estimate. To further improve the reliability of the estimate, utilize the multispectral information of the overall image to calculate the vegetation index value for each pixel. Based on a set threshold, binarize the vegetation index map to obtain a detailed vegetation distribution map. Combine the area estimate with the vegetation index distribution map and use a weighted average to obtain the final vegetation cover measurement result.

[0059] Step S106: To verify the accuracy of the vegetation cover measurement results, high-precision satellite remote sensing imagery is used as reference data. Geometric correction and registration are performed on the satellite imagery to ensure spatial consistency with the imagery from the handheld field device. Then, vegetation classification is performed on the satellite imagery to obtain a reference true value for vegetation cover. The measurement results from the handheld field device are compared with the reference true value at the pixel level, a confusion matrix is ​​calculated, and the accuracy of the measurement results is evaluated, including overall accuracy, user accuracy, and mapping accuracy, to ensure the reliability and practicality of the vegetation cover measurement method using the handheld field device.

[0060] Specifically:

[0061] A depth camera is installed on a handheld device in the field, and a heterogeneous depth image of the area to be tested is acquired based on the depth camera.

[0062] The deep heterogeneous images are preprocessed to obtain spatially consistent overall images, and the spatially consistent overall images are fused to obtain a fused dataset.

[0063] Furthermore, the process of preprocessing the depth heterogeneous image to obtain a spatially consistent overall image includes:

[0064] Significant feature points are extracted from the deep heterogeneous image, the similarity between the significant feature points is calculated, and geometric correction and spatial registration are performed on the deep heterogeneous image based on the similarity to obtain a spatially consistent overall image.

[0065] Furthermore, the process of fusing the spatially consistent overall images to obtain a fused dataset includes:

[0066] Extract feature points from the spatially consistent overall image and calculate the distribution density of the feature points;

[0067] The overlapping region of the image is determined based on the distribution density of the feature points, and the initial fusion weight of each pixel in the overlapping region is calculated.

[0068] For each pixel in the overlapping area, obtain its corresponding gray value in the image to be fused, and perform a weighted average of the gray values ​​according to the initial fusion weights to obtain the fused gray value of the pixel.

[0069] The fused pixel gray values ​​in the overlapping area are then combined with the original pixel gray values ​​in the non-overlapping area to obtain a preliminary fused image.

[0070] The preliminary fused images are subjected to image enhancement processing to generate the fused dataset.

[0071] Specifically, while handheld aerial photography in the field can capture vast amounts of surface information, a single image has a limited field of view. To obtain a complete regional image, multiple images taken with handheld devices need to be stitched together. Due to differences in shooting angle, altitude, and lighting, these images suffer from variations in field of view, resolution, and noise, necessitating a series of preprocessing and registration steps. First, each image undergoes preprocessing. For example, an image may contain random noise affecting its clarity. Gaussian filtering can be used to remove this noise. Suppose a pixel in an image has a grayscale value of 100, and its surrounding pixels have grayscale values ​​of 90, 110, 95, and 105. After Gaussian filtering, the pixel's grayscale value becomes a weighted average of these values, with the weights determined by a Gaussian function, thus smoothing the image and reducing noise. Furthermore, the image may be underexposed, resulting in low contrast. Histogram equalization can be used to enhance image contrast, making image details clearer. This is similar to adjusting the brightness and contrast of an image to make it "look better." Preprocessed images are of higher quality, which is beneficial for subsequent feature extraction. Next, salient feature points are extracted from the preprocessed image. For example, the SIFT algorithm is used to extract feature points. The SIFT algorithm identifies scale-invariant feature points in the image, such as corner points and edge points, which can still be identified even after the image is scaled or rotated.

[0072] Acquire multiple images to be fused, such as multiple photos of the same area taken from different locations using a handheld device in the field, with overlapping areas between them. Extract feature points from each image. The SIFT algorithm can be used to extract scale-invariant feature points, such as corner points and edge points. Calculate the feature point density, which can be done by counting the number of feature points per unit area, such as the number of feature points per square pixel. Regions with high feature point density are usually areas with rich texture in the image. Determine the overlapping regions based on the feature point density. Regions with high feature point density often correspond to overlapping areas. For example, regions with relatively dense feature points in both images can be considered overlapping regions. Calculate the initial fusion weight for each pixel within the overlapping region. The initial weight can be determined based on the distance of the pixel to the boundary of the overlapping region; the closer to the boundary, the smaller the weight, and the farther away, the larger the weight. For example, an inverse distance function can be used to calculate the weight, with pixels closer to the boundary having a weight close to 0 and pixels farther away having a weight close to 1. For each pixel within the overlapping region, obtain its corresponding grayscale value in the image to be fused. For example, a pixel has a grayscale value of 100 in the first image and a corresponding grayscale value of 150 in the second image. The grayscale values ​​are weighted and averaged according to the initial fusion weights to obtain the fused grayscale value of that pixel. Assuming the initial fusion weight for this pixel is 0.3 in the first image and 0.7 in the second image, then the fused grayscale value of this pixel is 0.3 × 100 + 0.7 × 150 = 135. It is then determined whether the difference between the fused grayscale value of the current pixel and its adjacent pixels exceeds a preset smoothing threshold. For example, if the preset smoothing threshold is 10, and the fused grayscale value of the current pixel is 135 while the fused grayscale value of its adjacent pixels is 150, then the difference is 15, exceeding the smoothing threshold. If it does, the fusion weights of the current pixel and its adjacent pixels are adjusted until the smooth transition condition is met. For example, the weight of the current pixel can be adjusted to 0.4, and the weight of its adjacent pixels to 0.6, and the fused grayscale value can be recalculated until the difference is less than or equal to 10. This is done to eliminate stitching artifacts and make the fused image look more natural and smooth. The weighted averaging and smoothing threshold determination steps are repeated until the fused grayscale values ​​of all pixels within the overlapping area meet the smooth transition condition, eliminating stitching artifacts. This is akin to fine-tuning an image, ensuring a smooth and natural transition in the overlapping area. The fused pixel grayscale values ​​within the overlapping area are then stitched together with the original pixel grayscale values ​​in the non-overlapping areas to obtain a preliminary fused image. The pixel grayscale values ​​in the non-overlapping areas are directly retained as their original values. For example, if the left side of the first image is a non-overlapping area, its original pixel grayscale values ​​are directly retained. Image enhancement processing is then applied to the preliminary fused image to improve the overall visual quality, outputting the final high-precision fused image.

[0073] A multi-feature fusion model for vegetation identification is constructed based on a random forest classifier. The fused dataset is input into the multi-feature fusion model for vegetation identification for calculation to obtain the binary segmentation result of the vegetation.

[0074] Furthermore, the process of obtaining the binary segmentation result of the vegetation includes:

[0075] After color conversion processing of the fused dataset, information extraction is performed to obtain color features, texture features, and shape features;

[0076] The color features, texture features, and shape features are normalized to obtain a multi-feature fusion vector;

[0077] A random forest classifier is constructed, and the random forest classifier is trained based on a historical dataset to obtain a multi-feature fusion model for vegetation recognition.

[0078] The multi-feature fusion vector is input into the multi-feature fusion model for vegetation recognition for calculation to obtain the binary segmentation result of the vegetation.

[0079] Furthermore, the process of extracting information from the fused dataset after color transformation to obtain color features, texture features, and shape features includes:

[0080] After converting the fused dataset from RGB space to HSV space, the green channel information is extracted to obtain the color features of green vegetation;

[0081] The texture features of the fused dataset are calculated based on the gray-level co-occurrence matrix method to obtain the texture features of the vegetation area;

[0082] Based on the shape characteristics of the vegetation area, extract the shape features of the vegetation.

[0083] The input color image is converted from RGB to HSV color space to extract green channel information, thus obtaining the color features of green vegetation. The gray-level co-occurrence matrix method is used to calculate the image's texture features, including contrast, correlation, energy, and entropy, to obtain the texture features of the vegetation region. Based on the shape characteristics of the vegetation region, morphological operations, such as opening and closing operations, are designed to extract the shape features of the vegetation and eliminate noise and small regions. Color, texture, and shape features are normalized to construct a multi-feature fusion vector, which serves as the input feature for a random forest classifier. The random forest classifier is used to train and predict the multi-feature fusion vector, and ensemble learning methods are employed to improve classification accuracy and robustness, resulting in a binary segmentation result for the vegetation. Post-processing of the binary segmentation result, such as morphological filtering and connected component analysis, is performed to eliminate isolated small regions and noise, obtaining the final vegetation segmentation mask. The vegetation segmentation mask is fused with the original image to highlight the vegetation region, achieving the extraction and visualization of green vegetation information.

[0084] Multi-source remote sensing imagery containing vegetated areas was acquired, and various effective vegetation feature vectors were extracted based on the characteristics of different data sources. The extracted multi-source features were optimized and combined using feature selection and feature transformation methods to obtain a fused high-dimensional feature vector representation. A random forest classifier was used to train the fused feature vector, and the generalization ability and robustness of the classifier were improved by ensembling multiple decision trees. During training, the parameters of the random forest, such as the number of decision trees, the minimum number of samples per node, and the size of the feature subset, were reasonably set according to the distribution of the sample data to balance classification accuracy and computational efficiency. The trained random forest classifier was used to predict new remote sensing images. The feature vector of each pixel was input into the classifier to obtain the probability value of whether the pixel belongs to vegetation or not. Based on the predicted vegetation probability map, a threshold was set to binarize it, resulting in a binary segmentation result of the vegetation.

[0085] The region merging method based on graph theory and the minimum spanning tree algorithm are used to process the binary segmentation results of the vegetation to obtain the merged vegetation region.

[0086] Furthermore, the process of obtaining the merged vegetation area includes:

[0087] The graph theory-based region merging method transforms the binary segmentation result of the vegetation into an undirected weighted graph. Through the minimum spanning tree algorithm, regions with high similarity are gradually merged until all regions are merged into a whole and then output, generating the merged vegetation region.

[0088] Specifically, in the process of extracting green vegetation information, constructing an undirected weighted graph is a crucial step in achieving vegetation region de-overlap. First, the vegetation regions in the extracted image are transformed into nodes in graph theory. Assume there are multiple independent vegetation regions in the image, each considered a node. For example, there are five salient vegetation regions in the image, labeled as nodes A, B, C, D, and E. The similarity between nodes is used as the weight of the edges to construct an undirected weighted graph. Similarity is calculated based on features such as color, texture, and shape. For example, nodes A and B have high similarity in color features but low similarity in texture and shape features. The similarity between nodes is measured using methods such as Euclidean distance or Mahalanobis distance. Assuming the Euclidean distance calculation shows that the color feature distance between nodes A and B is 2, the texture feature distance is 5, and the shape feature distance is 3, combining these distances, the similarity between nodes A and B is 4. The similarity between all nodes is calculated, forming a node similarity matrix. Each element in the matrix represents the similarity between a pair of nodes.

[0089] Based on the similarity matrix, the minimum spanning tree algorithm is used to progressively merge nodes with the highest similarity. Assuming initial merging of nodes A and B, resulting in a new node AB, the node features and similarity matrix are updated. The features of the new node AB are the weighted average of the features of nodes A and B, and the similarity matrix is ​​updated accordingly to reflect the similarity between the new node and other nodes. A similarity threshold is set to control the granularity of the merging process. Assuming a threshold of 5, the merging process stops when the similarity between nodes falls below 5.

[0090] Calculate the vegetation index value for each pixel in the merged vegetation region, generate a fine distribution map of vegetation based on the vegetation index value, and simultaneously calculate the vegetation coverage estimate of the merged vegetation region.

[0091] Furthermore, the process of generating a detailed vegetation distribution map based on the vegetation index values, and simultaneously calculating the estimated vegetation cover of the merged vegetation area, includes:

[0092] The vegetation area of ​​the merged vegetation region is obtained based on pixel statistics, the total proportion of the merged vegetation region is calculated, and a preliminary vegetation coverage estimate is obtained based on the total proportion.

[0093] Obtain the multispectral information of the overall image, and calculate the vegetation index value of each pixel based on the spectral information;

[0094] The NDVI value of the vegetation index for each pixel is calculated, and a vegetation index distribution map is generated based on the NDVI value.

[0095] The vegetation index distribution map is binarized to obtain a detailed vegetation distribution map.

[0096] For example, inputting the processed park image into a random forest classifier can effectively distinguish between vegetated and non-vegetated areas. Finally, the binary segmentation result output by the classifier is post-processed, such as by applying morphological filtering and connected component analysis, to eliminate isolated small regions and noise, resulting in a final vegetation segmentation mask. This mask can be fused with the original image to highlight vegetated areas, thereby achieving effective extraction and visualization of green vegetation information. This method has significant application value in fields such as urban greening monitoring and agricultural research, providing crucial information about vegetation coverage and health status.

[0097] The detailed distribution map of the vegetation and the estimated vegetation coverage are combined to generate the final vegetation coverage measurement result.

[0098] Furthermore, the process of generating the final vegetation cover measurement result includes:

[0099] The number of pixels with a value of 1 in the detailed distribution map of the vegetation is counted, and then divided by the total number of pixels in the image to obtain the detailed vegetation coverage rate based on the vegetation index.

[0100] The final vegetation coverage measurement result is obtained by weighting the preliminary vegetation coverage estimate and the fine vegetation coverage based on the vegetation index.

[0101] The merged vegetation region image is acquired, and the area of ​​the vegetation region is obtained through pixel statistics. The proportion of the vegetation region to the total image area is calculated to obtain a preliminary estimate of vegetation coverage. Multispectral information of the overall image is acquired, and the Normalized Difference Vegetation Index (NDVI) value is calculated for each pixel location to obtain a vegetation index distribution map. Based on a preset NDVI threshold, the vegetation index distribution map is binarized, with pixels less than the threshold set to 0 and pixels greater than or equal to the threshold set to 1, resulting in a refined binarized distribution map of the vegetation. The number of pixels with a value of 1 in the binarized distribution map is counted and divided by the total number of pixels in the image to obtain a result based on... The fine vegetation cover rate is obtained by calculating the vegetation index; a weighted average is calculated between the preliminary area estimate and the fine vegetation index coverage rate, with the weighting coefficients determined based on the reliability of the two methods, to obtain the final vegetation cover rate measurement result; a decision tree algorithm is used, with vegetation index and spectral information as features and coverage rate as the objective, to train a regression prediction model for rapid estimation of vegetation cover rate in unknown areas; for continuous multi-temporal remote sensing images, the vegetation cover rate of each temporal phase is extracted, and the dynamic trend of vegetation cover change is obtained through time series analysis, further analyzing its correlation with environmental factors such as climate and topography.

[0102] Assuming an image has a total pixel count of 1000×1000, after region merging, the vegetation area has 300,000 pixels. Therefore, the vegetation area accounts for 30% of the entire image, which is the preliminary estimate of vegetation cover. This method is simple and intuitive, but may be affected by image resolution and segmentation accuracy. To improve the estimation accuracy of vegetation cover, multispectral information of the entire image is further obtained. Multispectral images contain reflectance data from different bands, reflecting the spectral characteristics of vegetation. For each pixel location, its Normalized Difference Vegetation Index (NDVI) value is calculated. The formula for NDVI is: (NIR-Red) / (NIR+Red), where NIR represents the reflectance in the near-infrared band, and Red represents the reflectance in the red band. Through calculation, an NDVI distribution map can be obtained, where the value of each pixel represents the vegetation health status at that location. The NDVI distribution map is then binarized according to a preset NDVI threshold. Assuming a threshold of 4, pixel values ​​less than 4 are set to 0, representing non-vegetated areas; pixel values ​​greater than or equal to 4 are set to 1, representing vegetated areas. This process yields a refined binary distribution map of vegetation. The number of pixels with a value of 1 is counted, assuming it's 350,000. Dividing this number by the total number of pixels in the image (1,000,000) gives a refined vegetation coverage rate of 35% based on the vegetation index. This method utilizes the spectral characteristics of vegetation to more accurately reflect its actual distribution.

[0103] Example 2

[0104] This embodiment provides a vegetation coverage measurement system, including:

[0105] The data acquisition module is used to install a depth camera on a handheld device in the field and acquire a heterogeneous depth image of the area to be tested based on the depth camera.

[0106] The data fusion module is used to preprocess the deep heterogeneous images to obtain spatially consistent overall images, and to fuse the spatially consistent overall images to obtain a fused dataset.

[0107] The model calculation module is used to construct a multi-feature fusion model for vegetation recognition based on a random forest classifier. The fusion dataset is input into the multi-feature fusion model for vegetation recognition for calculation to obtain the binary segmentation result of the vegetation.

[0108] The merging module is used to process the binary segmentation results of the vegetation based on graph theory-based region merging methods and minimum spanning tree algorithms to obtain merged vegetation regions.

[0109] The estimation calculation module is used to calculate the vegetation index value of each pixel in the merged vegetation area, generate a fine distribution map of vegetation based on the vegetation index value, and calculate the vegetation coverage estimate of the merged vegetation area.

[0110] The result generation module is used to combine the detailed distribution map of the vegetation with the vegetation coverage estimate to generate the final vegetation coverage measurement result.

[0111] The above are merely preferred embodiments of this application, but 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 be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for measuring vegetation cover, characterized in that, Includes the following steps: A depth camera is installed on a handheld device in the field, and a heterogeneous depth image of the area to be tested is acquired based on the depth camera. The deep heterogeneous images are preprocessed to obtain spatially consistent overall images, and the spatially consistent overall images are fused to obtain a fused dataset. A multi-feature fusion model for vegetation identification is constructed based on a random forest classifier. The fused dataset is input into the multi-feature fusion model for vegetation identification for calculation to obtain the binary segmentation result of the vegetation. The region merging method based on graph theory and the minimum spanning tree algorithm are used to process the binary segmentation results of the vegetation to obtain the merged vegetation region. Calculate the vegetation index value for each pixel in the merged vegetation region, generate a fine distribution map of vegetation based on the vegetation index value, and simultaneously calculate the vegetation coverage estimate of the merged vegetation region. The detailed distribution map of the vegetation and the estimated vegetation coverage are combined to generate the final vegetation coverage measurement result. During the image fusion process, the fusion weights are adaptively adjusted according to the distribution of feature points in the overlapping areas of the images, and the pixel gray values ​​in the overlapping areas are weighted and averaged to ensure that the fused image transitions smoothly in the overlapping areas. Using high-precision satellite remote sensing imagery as reference data, the satellite imagery is geometrically corrected and registered to make it spatially consistent with the imagery from handheld field devices. Vegetation classification is performed on satellite imagery to obtain a reference true value of vegetation cover; the measurement results from handheld field devices are compared with the reference true value at the pixel level, the confusion matrix is ​​calculated, and the accuracy of the measurement results is evaluated; The adaptive adjustment of fusion weights includes: acquiring multiple images to be fused; using the SIFT algorithm to extract scale-invariant feature points from the images; calculating the distribution density of the feature points; determining the overlapping region of the images based on the feature point distribution density; calculating the initial fusion weight of each pixel in the overlapping region, where the initial weight is determined based on the distance from the pixel to the boundary of the overlapping region (the closer to the boundary, the smaller the weight; the farther from the boundary, the larger the weight); for each pixel in the overlapping region, acquiring its corresponding gray value in the image to be fused; performing a weighted average of the gray values ​​based on the initial fusion weight to obtain the fused gray value of the pixel; determining whether the difference between the fused gray values ​​of the current pixel and its neighboring pixels exceeds a preset smoothing threshold; if it does, adjusting the fusion weights of the current pixel and its neighboring pixels until the smooth transition condition is met; repeating the weighted average and smoothing threshold determination steps until the fused gray values ​​of all pixels in the overlapping region meet the smooth transition condition and the stitching marks are eliminated; stitching the fused pixel gray values ​​in the overlapping region with the original pixel gray values ​​in the non-overlapping region to obtain a preliminary fused image; performing image enhancement processing on the preliminary fused image; and outputting the final high-precision fused image. The process of preprocessing the depth heterogeneous image to obtain a spatially consistent overall image includes: Significant feature points are extracted from the deep heterogeneous image, the similarity between the significant feature points is calculated, and geometric correction and spatial registration are performed on the deep heterogeneous image based on the similarity to obtain a spatially consistent overall image.

2. The method for measuring vegetation coverage according to claim 1, characterized in that, The process of obtaining the binary segmentation result of vegetation includes: After color conversion processing of the fused dataset, information extraction is performed to obtain color features, texture features, and shape features; The color features, texture features, and shape features are normalized to obtain a multi-feature fusion vector; A random forest classifier is constructed, and the random forest classifier is trained based on a historical dataset to obtain a multi-feature fusion model for vegetation recognition. The multi-feature fusion vector is input into the multi-feature fusion model for vegetation recognition for calculation to obtain the binary segmentation result of the vegetation.

3. The method for measuring vegetation coverage according to claim 2, characterized in that, The process of extracting information from the fused dataset after color transformation to obtain color features, texture features, and shape features includes: After converting the fused dataset from RGB space to HSV space, the green channel information is extracted to obtain the color features of green vegetation; The texture features of the fused dataset are calculated based on the gray-level co-occurrence matrix method to obtain the texture features of the vegetation area; Based on the shape characteristics of the vegetation area, extract the shape features of the vegetation.

4. The method for measuring vegetation coverage according to claim 1, characterized in that, The process of obtaining the merged vegetation area includes: The graph theory-based region merging method transforms the binary segmentation result of the vegetation into an undirected weighted graph. Through the minimum spanning tree algorithm, regions with high similarity are gradually merged until all regions are merged into a whole and then output, generating the merged vegetation region.

5. The method for measuring vegetation coverage according to claim 1, characterized in that, The process of generating a detailed vegetation distribution map based on the vegetation index values, and simultaneously calculating the estimated vegetation cover of the merged vegetation area, includes: The vegetation area of ​​the merged vegetation region is obtained based on pixel statistics, the total proportion of the merged vegetation region is calculated, and a preliminary vegetation coverage estimate is obtained based on the total proportion. Obtain the multispectral information of the overall image, and calculate the vegetation index value of each pixel based on the spectral information; The NDVI value of the vegetation index for each pixel is calculated, and a vegetation index distribution map is generated based on the NDVI value. The vegetation index distribution map is binarized to obtain a detailed vegetation distribution map.

6. The method for measuring vegetation coverage according to claim 5, characterized in that, The process of generating the final vegetation cover measurement results includes: The number of pixels with a value of 1 in the detailed distribution map of the vegetation is counted, and then divided by the total number of pixels in the image to obtain the detailed vegetation coverage rate based on the vegetation index. The final vegetation coverage measurement result is obtained by weighting the preliminary vegetation coverage estimate and the fine vegetation coverage based on the vegetation index.

7. The system for measuring vegetation cover according to any one of claims 1-6, characterized in that, include: The data acquisition module is used to install a depth camera on a handheld device in the field and acquire a heterogeneous depth image of the area to be tested based on the depth camera; The data fusion module is used to preprocess the deep heterogeneous images to obtain spatially consistent overall images, and to fuse the spatially consistent overall images to obtain a fused dataset. The model calculation module is used to construct a multi-feature fusion model for vegetation recognition based on a random forest classifier. The fusion dataset is input into the multi-feature fusion model for vegetation recognition for calculation to obtain the binary segmentation result of the vegetation. The merging module is used to process the binary segmentation results of the vegetation based on graph theory-based region merging methods and minimum spanning tree algorithms to obtain merged vegetation regions. The estimation calculation module is used to calculate the vegetation index value of each pixel in the merged vegetation area, generate a fine distribution map of vegetation based on the vegetation index value, and calculate the vegetation coverage estimate of the merged vegetation area. The result generation module is used to combine the detailed distribution map of the vegetation with the vegetation coverage estimate to generate the final vegetation coverage measurement result.

Images