A method for detecting vegetation in mining areas based on satellite remote sensing images

Abstract

The present invention relates to the field of image processing technology, and specifically relates to a method for detecting mine area vegetation based on satellite remote sensing images. The detection method includes: first, collecting remote sensing images of the mine area and evaluating the feature complexity of the remote sensing images and their respective channel images; then, calculating the importance degree of each channel according to the feature complexity and contrast of the remote sensing images and their respective channel images; subsequently, extracting the feature vectors of the remote sensing images and their respective channel images, and calculating the weights of each channel in combination with the importance degree of each channel; then, updating the pixel values of each pixel point in the remote sensing image according to the weights of each channel to obtain an updated remote sensing image; finally, using the OTSU algorithm to segment the updated remote sensing image to obtain the vegetation area. This method improves the accuracy of the image segmentation result and realizes the accurate detection of the mine area vegetation area.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing. Specifically, it relates to a method for detecting mining area vegetation based on satellite remote sensing images. Background Art

[0002] While obtaining resources, mining activities inevitably have a negative impact on the ecological environment around the mining area, and the damage to vegetation is a key issue. Since the growth condition of vegetation is an important indicator of the ecological environment quality, accurately detecting the vegetation in the mining area is of great significance for evaluating the degree of ecological environment damage.

[0003] The prior art usually collects remote sensing images of the mining area, then segments the remote sensing images, and obtains the vegetation area according to the segmentation results to achieve vegetation detection. The OTSU algorithm is a common image segmentation method. For example, the Chinese patent document with the publication number CN107705283B discloses a method for detecting particle and bubble collisions based on OTSU image segmentation. The method includes: first, optimizing the image by using the preprocessing operation of the image, and segmenting the bubbles and particles through the OTSU algorithm; then, fitting the particles and bubbles into circles by combining the least squares method; finally, tracking the positions of all the particles and detecting whether the particles and bubbles collide.

[0004] However, when using the above OTSU algorithm to segment the remote sensing images of the mining area, considering that a large area of the mining area is covered by vegetation, observing from the gray level histogram of the remote sensing image, it usually does not present two clearly distinguishable obvious peaks, that is, it presents a non-bimodal form. For such remote sensing images with a non-bimodal gray level histogram, the OTSU algorithm is difficult to accurately determine the threshold for distinguishing vegetation from other parts during segmentation, which affects the segmentation effect. This makes the algorithm unable to accurately and completely segment the vegetation area from the entire remote sensing image, affecting the accuracy and reliability of the vegetation detection results. Summary of the Invention

[0005] To solve the problem that due to the existence of a large area of complex vegetation in the remote sensing images of the mining area, the gray level histogram of the remote sensing image presents a non-bimodal form, and the OTSU algorithm has a poor segmentation effect on the remote sensing image, affecting the accuracy and reliability of the vegetation detection results, the present invention proposes a method for detecting mining area vegetation based on satellite remote sensing images, including:

[0006] Collecting remote sensing images of the mining area and evaluating the feature complexity of the remote sensing image and its channel images;

[0007] Calculating the importance degree of each channel according to the feature complexity of the remote sensing image and its channel images, and the contrast of each channel image , where is the importance degree of the th channel is the feature complexity of the remote sensing image, is the feature complexity of the th channel image of the remote sensing image, is the th channel image contrast of the remote sensing image, is the linear normalization function;

[0008] Extract the feature vectors of the remote sensing image and its channel images, and calculate the weights of each channel of the remote sensing image by combining the importance of each channel: , where is the weight of the th channel of the remote sensing image, is the cosine similarity between the feature vector of the remote sensing image and the feature vector of the th channel image, is the natural exponential function, is the total number of channels;

[0009] For the pixel points in the remote sensing image, update the pixel values of the pixel points according to the weights of each channel to obtain the updated remote sensing image. Segment the updated remote sensing image by the OTSU algorithm, and determine the vegetation area according to the segmentation result to achieve vegetation detection in the mining area.

[0010] The above technical solution can quantitatively describe the remote sensing image and its channel images from multiple dimensions by evaluating the feature complexity of the remote sensing image and its channel images, which helps to distinguish different types of ground objects. For the vegetation detection in mining areas, there are differences in the feature complexity among different types of vegetation and other ground objects (such as bare land, mining facilities, etc.). And further, by comprehensively considering the feature complexity and contrast of the channel images, the importance of each channel is calculated. The feature complexity reflects the complex characteristics of the image, while the contrast can reflect the difference degree between different regions in the image. Combining the two can more comprehensively measure the importance of each channel of the remote sensing image for vegetation detection. And further, by calculating the cosine similarity between the feature vectors of the remote sensing image and the feature vectors of the channel images, the similarity between them can be measured. Combining the previously calculated importance to calculate the weight of each channel, the weight can accurately reflect the relative importance of each channel in the final vegetation detection, and can more accurately assign weights to different channels, improving the accuracy of the final detection result and having better adaptability for vegetation detection in complex mining environments. And further, by updating the pixel points according to the calculated weight, the importance and feature similarity information of different channels obtained from the previous analysis can be fused into the pixel values of the image, forming an updated remote sensing image, so that the image contains information more conducive to vegetation detection. The weighted updated remote sensing image contains information comprehensively considering the importance and feature similarity of each channel. Compared with the original image or simply processed image, it is more conducive to highlighting the vegetation area and improving the accuracy of vegetation detection. The OTSU threshold segmentation method can find the optimal segmentation threshold according to the updated image, overcoming the deficiency of the original OTSU algorithm when processing mining area remote sensing images (the gray histogram is non-bimodal), more accurately segmenting the vegetation area, improving the reliability and accuracy of vegetation detection, and finally realizing the effective detection of mining area vegetation, providing more reliable data support for evaluating the damage degree of the mining area ecological environment, etc.

[0011] Preferably, the method for evaluating the feature complexity of the remote sensing image and its channel images is as follows:

[0012] For any image in the remote sensing image and its channel images, use the local binary pattern to obtain the feature value of each pixel point in the image, construct a histogram of the feature values, the abscissa of the histogram is different feature values, and the ordinate is the number of pixel points corresponding to each feature value;

[0013] Calculate the feature complexity of the image: ; where is the feature complexity of the image, is the variance of the number of pixel points corresponding to all feature values in the histogram, is the maximum value of the number of pixel points corresponding to all feature values in the histogram, is the number of eigenvalues that have appeared in the image, is the total number of eigenvalues.

[0014] The above technical solution can more comprehensively and meticulously reflect the complexity of the texture of the image by comprehensively analyzing information in multiple dimensions of the image.

[0015] Preferably, a method for updating the pixel value of a pixel point according to the weights of each channel is as follows:

[0016] Directly perform weighted summation on the pixel values of the pixel point in each channel according to the weights of each channel to achieve the update of the pixel value of the pixel point.

[0017] The above technical solution helps to improve the integrity and accuracy of image information. By integrating information from different channels, it can more accurately represent the information of pixel points and provide a better-quality input for subsequent OTSU threshold segmentation.

[0018] Preferably, another method for updating the pixel value of a pixel point according to the weights of each channel is as follows:

[0019] First, if the weight of a certain channel is the smallest, set the weight of this channel to 0;

[0020] Then, perform weighted summation on the pixel values of the pixel point in each channel according to the weights of each channel to achieve the update of the pixel value of the pixel point.

[0021] The above technical solution can reduce unnecessary computational complexity by excluding the channel with the smallest weight, and at the same time reduce the noise or error information brought by low-value or interfering channels, making the finally updated pixel value more accurately reflect the characteristics of the vegetation area.

[0022] Preferably, the method for determining the vegetation area according to the segmentation result is as follows:

[0023] Take the pixel points with pixel value 0 in the segmentation result of the updated remote sensing image as vegetation pixel points, and the area composed of all vegetation pixel points as the vegetation area.

[0024] Preferably, the method for extracting the feature vectors of the remote sensing image and its channel images is as follows:

[0025] For any one of the remote sensing image and its channel images, construct a histogram of the pixel values of this image. The abscissa of the histogram is different pixel values, and the ordinate is the number of pixel points corresponding to each pixel value. The number of pixel points corresponding to all pixel values in this image constitutes the feature vector of this image.

[0026] In the above technical solution, the number of pixel points corresponding to all pixel values in the image is used as an element of the feature vector. This method is simple and direct, and can effectively capture the overall feature information of the image.

[0027] Preferably, the contrast of each channel image is any one of Weber contrast, Michelson contrast, and root mean square contrast.

[0028] Preferably, each channel is a red channel, a green channel, and a blue channel respectively.

[0029] Preferably, the vegetation detection further includes:

[0030] After determining the vegetation area, the ratio of the number of pixel points in the vegetation area to the number of all pixel points in the updated remote sensing image is used as the vegetation coverage rate of the mining area.

[0031] The above technical solution can intuitively reflect the proportion of the vegetation in the mining area in the whole region in a quantitative way, which is of great significance for comprehensively understanding the current situation of the ecological environment in the mining area and detecting its change trend over time.

[0032] The present invention has the following effects:

[0033] The present invention analyzes the remote sensing image from multiple dimensions through a unique method, updates the pixel values of the pixel points in the remote sensing image, makes the updated image more prominent in the vegetation area, and uses the OTSU algorithm to accurately find the segmentation threshold based on the updated remote sensing image, overcoming the defects of the traditional OTSU algorithm in segmenting the remote sensing image, improving the accuracy of image segmentation, accurately segmenting the vegetation area in the remote sensing image, and making the vegetation detection result in the mining area more reliable. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] By referring to the accompanying drawings and reading the following detailed description, the above and other objects, features, and advantages of the exemplary embodiments of the present invention will become readily understood. In the drawings, several embodiments of the present invention are shown in an exemplary rather than restrictive manner, and the same or corresponding reference numerals represent the same or corresponding parts, wherein:

[0035] Figure 1 is a schematic flowchart of the method of the present invention;

[0036] Figure 2 is the remote sensing image of the present invention;

[0037] Figure 3 is the grayscale histogram of the remote sensing image of the present invention;

[0038] Figure 4 is the initial segmentation effect diagram of the remote sensing image of the present invention;

[0039] Figure 5 is the histogram of the eigenvalue of the remote sensing image of the present invention;

[0040] Figure 6 is the blue channel map of the remote sensing image of the present invention;

[0041] Figure 7 is the green channel map of the remote sensing image of the present invention;

[0042] Figure 8 is the red channel map of the remote sensing image of the present invention;

[0043] Figure 9 is the segmentation effect diagram of the updated remote sensing image of the present invention. Detailed implementation manners

[0044] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative efforts fall within the protection scope of the present invention.

[0045] Next, the detailed implementation manners of the present invention will be described in detail in conjunction with the accompanying drawings.

[0046] Refer to Figure 1 , a method for detecting mine area vegetation based on satellite remote sensing images provided by the present invention includes steps S1 - S7:

[0047] S1: Collect remote sensing images of the mine area.

[0048] An unmanned aerial vehicle (UAV) is a flexible and cost - effective remote sensing platform that can carry cameras or sensors and is usually used for high - resolution data collection in small - scale areas. Therefore, in this step, the UAV carries a multispectral camera to collect remote sensing images of the mine area. The multispectral camera is a key device for obtaining remote sensing images and is commonly used in fields such as crop growth assessment, coverage assessment, and water quality detection. The multispectral camera can simultaneously obtain image information in multiple different bands, such as the common red, green, blue, and near - infrared bands. Different ground objects have different reflection characteristics in these bands. For example, vegetation has a high reflectance in the near - infrared band. By recording these differences with the multispectral camera, rich data can be provided for subsequent analysis, thereby achieving accurate detection of vegetation coverage.

[0049] As Figure 2 shown, the vegetation in the collected remote sensing images is widely distributed and in a complex situation. As Figure 3 shown, the gray - level histogram of the remote sensing image presents an obvious non - bimodal shape. As Figure 4As shown in [figure], the segmentation effect of the remote sensing image using the OTSU algorithm is poor. The black part represents the vegetation area, and the white part represents other ground objects. It is difficult to accurately distinguish vegetation from other ground objects. This indicates that for such remote sensing images with non-bimodal gray histograms, the OTSU algorithm has certain limitations and cannot meet the requirement of precise segmentation of vegetation distribution.

[0050] Therefore, in the subsequent steps, the present invention first obtains the channel images of the remote sensing image. The remote sensing image is an image after grayscale processing. The pixel value of each pixel point in the remote sensing image is the gray value of that pixel point. The pixel value of each pixel point in each channel image of the remote sensing image is equal to the channel value of that pixel point. For example, in the red channel image, the pixel value of each pixel point is the red channel value, and the same applies to the other channel images.

[0051] Then, by analyzing the feature complexity of the remote sensing image and its channel images, according to the feature complexity of the remote sensing image and its channel images, and the contrast of each channel image, calculate the importance degree of each channel, extract the feature vectors of the remote sensing image and its channel images, and combine the importance degree of each channel to calculate the weights of each channel of the remote sensing image. For the pixel points in the remote sensing image, update the pixel values of the pixel points according to the weights of each channel to obtain the updated remote sensing image. By Threshold segment the updated remote sensing image, and determine the vegetation area according to the segmentation result, improving the segmentation effect to achieve accurate vegetation detection.

[0052] S2: Evaluate the feature complexity of the remote sensing image and its channel images.

[0053] Generally speaking, because vegetation and open spaces in remote sensing images have different characteristics, image segmentation algorithms can be used to segment vegetation from the images to achieve vegetation detection. However, the actual situation is more complex. In the mining area environment, there are often multiple plants in the same area, and the growth states of these plants are different. This results in different pixel point characteristics corresponding to different plants in the remote sensing image. For example, healthy growing green plants and plants affected by pests and diseases have differences in pixel characteristics such as color and texture in the remote sensing image, increasing the complexity of segmenting the remote sensing image.

[0054] Considering the above, the present invention quantifies the complexity of the remote sensing image. The greater the complexity, the more features of the pixel points the remote sensing image contains, and the more likely the remote sensing image contains multiple different states of vegetation.

[0055] In one embodiment, for any one of the remote sensing image and its channel images, the complexity can be evaluated according to the following method:

[0056] Use local binary pattern to obtain the eigenvalue of each pixel in the image, construct a histogram of eigenvalues, where the abscissa of the histogram is different eigenvalues and the ordinate is the number of pixels corresponding to each eigenvalue.

[0057] Then the complexity of the image is:

[0058]

[0059] In this formula, is the feature complexity of the image, is the variance of the number of pixels corresponding to all eigenvalues in the histogram, that is, the variance of the ordinates corresponding to all eigenvalues (abscissa) of the histogram, is the maximum value of the number of pixels corresponding to all eigenvalues in the histogram, is the number of eigenvalues that have appeared in the remote sensing image, is the total number of eigenvalues, .

[0060] In one embodiment, the method for evaluating the feature complexity of a remote sensing image is:

[0061] First, use local binary pattern to obtain the eigenvalue of each pixel in the remote sensing image. Specifically, the local binary pattern under eight-point sampling is adopted. This method can effectively capture the local texture features of the remote sensing image. Different texture structures correspond to different eigenvalues, providing rich basic data for subsequent analysis. Implementers can flexibly select the sampling method according to the actual situation to meet the requirements of image feature extraction in different scenarios.

[0062] Then, count the eigenvalues of all pixels in the remote sensing image, and construct a histogram of eigenvalues according to the eigenvalues of all pixels. As Figure 5 shown, the abscissa of the histogram is different eigenvalues, and the ordinate is the frequency of each eigenvalue appearing in the remote sensing image, that is, the number of pixels corresponding to each eigenvalue. For example, the ordinate at the abscissa of 20 in the histogram is 10, indicating that in the remote sensing image, there are 20 pixels with an eigenvalue of 20. The histogram intuitively shows the distribution of eigenvalues in the remote sensing image, providing a key basis for further analyzing the feature complexity of the remote sensing image.

[0063] Finally, calculate the feature complexity of the remote sensing image based on the information of the above histogram:

[0064]

[0065] In this formula, is the feature complexity of the remote sensing image, It is the variance of the number of pixel points corresponding to all the eigenvalues in the histogram, that is, the variance of the ordinates corresponding to all the eigenvalues (abscissas) of the histogram. It is the maximum value of the number of pixel points corresponding to all the eigenvalues in the histogram. It is the number of eigenvalues that have appeared in the remote sensing image. It is the total number of eigenvalues. Since eight-point sampling is used here, the total number of eigenvalues .

[0066] For in this formula:

[0067] When is relatively small, it means that the number of occurrences of each eigenvalue in the eigenvalue histogram is relatively close, indicating that in the remote sensing image, the distribution of the eigenvalues of pixel points is more dispersed, which implies that there may be multiple different local features in the remote sensing image. For example, there are various types of vegetation, different-shaped mine pits, exposed rocks and soils, etc. intertwined in the remote sensing image, presenting complex and diverse textures, making the feature complexity of the remote sensing image relatively large.

[0068] On the contrary, when is relatively large, it means that the number of occurrences of each eigenvalue in the eigenvalue histogram varies greatly, which implies that there may be prominent and relatively single local features in the remote sensing image. For example, a large area of single vegetation coverage area or a vast flat mining area, making the feature complexity of the remote sensing image relatively small.

[0069] For in this formula: Since the eigenvalues of the local binary pattern represent the local features of the image, so the larger it is, the fewer cases where different pixel points in the remote sensing image have the same eigenvalue, which means that the remote sensing image contains more diverse local features, and the feature complexity of the remote sensing image is also greater. On the contrary, the smaller it is, the more cases where different pixel points in the remote sensing image have the same eigenvalue, which means that the remote sensing image contains fewer local features. Therefore, the feature complexity of the remote sensing image is also smaller. And by dividing by , the normalization operation of is realized, so that the value range of is standardized to the interval, which is convenient for comparing the feature complexity between different images.

[0070] For : It is a key reference point for the frequency distribution of eigenvalues in the histogram. Essentially, it reflects which eigenvalue corresponds to the largest number of pixel points among all different eigenvalues, that is, the frequency of occurrence of the dominant local features in the remote sensing image. For example, in a remote sensing image, if the eigenvalues of certain pixel points are very large, it means that the local features represented by this eigenvalue are widely present in the remote sensing image, indicating that these pixel points are large-area regions in the remote sensing image, such as large areas of specific vegetation textures or certain common minefield structures.

[0071] For in this formula: This proportional relationship affects the calculation result of feature complexity. When is small and is relatively large, the value of will be small, which indicates that although there is a dominant local feature in the remote sensing image (the larger is), the distribution of other eigenvalues is also relatively uniform (the smaller is). At this time, the value of is closer to 1, and the finally calculated feature complexity will be relatively large, meaning that there are various different types of ground objects in the remote sensing image. Although there is a ground object that has an area advantage, the advantage is not absolute, and other ground objects are also widely present and evenly distributed. For example, in a remote sensing image of a mining area, there may be areas of dense forest vegetation, exposed rock areas, reclaimed farmland areas, and some small mine pit buildings, etc., resulting in a relatively high feature complexity of the remote sensing image.

[0072] Conversely, if is large and is relatively small, the value of will be large, which indicates that there is a relatively unprominent local feature in the remote sensing image (the smaller is), but the distribution of other eigenvalues has a large difference (the larger is). At this time, the value of is closer to 0, and the finally calculated feature complexity will be relatively small, meaning that there is no prominent dominant local feature in the remote sensing image and the distribution of other eigenvalues has a large difference. For example, in some small-scale mining areas in the remote sensing image, there are a small amount of vegetation, scattered rocks, and small mine pits, etc. Although the distribution of their eigenvalues has a large difference, the number of pixel points corresponding to the eigenvalues of each ground object is not large, resulting in a relatively low feature complexity of the remote sensing image.

[0073] In summary, the above technical solution can accurately quantify the feature complexity of the remote sensing image. The greater the feature complexity, it means that there are various different types of ground objects in the remote sensing image. On the contrary, the smaller the feature complexity, it means that the types of ground objects in the remote sensing image are relatively simple.

[0074] Furthermore, the present invention also takes into account that under normal circumstances, the various colors in any image are composed of the pixel values of the red channel, green channel, and blue channel, and the pixel values of different channels have different expressions for different colors. Therefore, after obtaining the channel images of the remote sensing image such as Figure 6 , Figure 7 , Figure 8 shown in, and combining Figure 6 and Figure 8 it can be seen that in the remote sensing image, the pixel points of vegetation and the pixel points of soil are very close in the pixel values of the blue channel, while there are obvious differences between the pixel points of vegetation and the pixel points of soil in the red channel. Therefore, this difference can be used to assist in vegetation detection.

[0075] First, obtain the feature complexity of the channel images of the remote sensing image (the same as the method for obtaining the feature complexity of the remote sensing image). Taking the red channel image as an example: First, use the local binary pattern to obtain the feature value of each pixel point in the red channel image. Specifically, the local binary pattern under eight-point sampling is adopted. Then, count the feature values of all pixel points in the red channel image, and construct a histogram of the feature values based on the feature values of all pixel points. The abscissa of the histogram is different feature values, and the ordinate is the number of pixel points corresponding to each feature value, that is, the frequency of each feature value appearing in the red channel image. Finally, calculate the feature complexity of the red channel image based on the information of this histogram. The calculation formula is the same as the calculation formula for the feature complexity of the remote sensing image.

[0076] S3: Calculate the importance degree of each channel according to the feature complexity of the remote sensing image and its channel images, and the contrast of each channel image.

[0077] Contrast is the degree of difference between the bright part and the dark part in an image. A channel image with high contrast usually means that the difference between the bright part and the dark part in this channel image is large, indicating that this channel contains more detailed and variable information, while low contrast is relatively plain, with a small difference between the bright part and the dark part, and is more likely to imply that the information contained in this channel image is less, and it is more difficult to extract useful features from it.

[0078] In remote sensing images, for different types of ground objects, their contrast performances in different channels are different. For example, vegetation may have a relatively high contrast in the near-infrared channel because the reflectance of vegetation has unique characteristics in the near-infrared band, which is significantly different from that of surrounding soil, water bodies, etc., thus forming a relatively high contrast. This high contrast helps to identify and distinguish vegetation from other ground objects. However, some ground objects may have a relatively low contrast in certain channels, which may mean that their characteristics in that channel are not prominent and it is difficult to distinguish them from other ground objects. In short, calculating the contrast helps to determine which channels are more conducive to the extraction and identification of ground object characteristics.

[0079] In one embodiment, the method for obtaining the contrast of each channel image can be arbitrarily selected from Weber contrast, Michelson contrast, and root mean square contrast. Here, Weber contrast is selected.

[0080] Weber contrast is based on the brightness of pixel points in the image and measures the contrast by calculating the relative difference in brightness between adjacent regions. It can sensitively capture the subtle brightness changes in the image and is suitable for scenarios with high requirements for the contrast of image details. For example, when distinguishing the subtle texture differences between vegetation and surrounding soil in a remote sensing image, Weber contrast can achieve good results and accurately present the detail contrast between the two; Michelson contrast focuses on the overall brightness distribution of the image and performs well in measuring the overall light and dark contrast difference of the image, being able to intuitively reflect the relative proportional relationship between the brightest and darkest parts of the image. For example, when analyzing the contrast between a large area of water and land in a remote sensing image, this method can clearly present the significant light and dark contrast between the two, which is of great significance for distinguishing macroscopic ground object characteristics; Root mean square contrast starts from a statistical perspective and quantifies the degree of dispersion of the brightness values of pixel points in the image. This method is sensitive to the comprehensive changes in brightness in the image and can reflect the overall contrast characteristics of the image. When analyzing a remote sensing image containing multiple types of ground objects with complex brightness changes, root mean square contrast can provide comprehensive contrast information and help to grasp the overall characteristics of the image.

[0081] Furthermore, the method for obtaining the contrast of the remote sensing image is also arbitrarily selected from Weber contrast, Michelson contrast, and root mean square contrast. Here, Weber contrast is selected.

[0082] In one embodiment, the importance levels of each channel satisfy the following relational expression:

[0083]

[0084] In the formula, is the importance level of the th channel, is the feature complexity of the remote sensing image, is the The feature complexity of the th channel image of the remote sensing image, is the contrast (Weber contrast) of the th channel image of the remote sensing image, is the linear normalization function,

[0085] In this formula, is the contrast of the th channel image of the remote sensing image. The larger this value is, the stronger the contrast feature of the th channel of the remote sensing image. The th channel is more likely to clearly reflect the difference between vegetation and other ground objects. In the vegetation detection task, the th channel is more likely to provide key information, and the th channel is more important. The smaller it is, the weaker the contrast feature of the th channel of the remote sensing image. The th channel is more difficult to highlight the difference between vegetation and other ground objects. In the vegetation detection task, the th channel is less able to provide key information, and the th channel is less important.

[0086] In this formula, The smaller it is, the lower the feature complexity of the th channel image, indicating that the texture features in this channel image are relatively single and it is easier to highlight the difference between vegetation and other ground objects. Therefore, the th channel is more important for the detection of vegetation areas, and the corresponding importance is also greater. On the contrary, The larger it is, the higher the feature complexity of the th channel image, indicating that the texture features in this channel image are relatively complex and diverse, and it is more difficult to clearly distinguish the vegetation area from other ground objects, thus reducing the importance of this channel for vegetation detection, and the corresponding importance is also smaller. Therefore, is used to correct by gamma transformation, specifically:

[0087] When is smaller, is smaller, then is closer to 0. At this time, has an upward correction effect on , making the th channel more important. For example, when the Contrast of one channel When it is at a medium level, but due to is very small, that is, the features are simple, and it is easier to distinguish the vegetation area from other ground objects. By upward correction, the importance calculated based on the contrast of this channel is increased, which is consistent with the actual situation that this channel is more valuable in vegetation detection.

[0088] When is larger, is larger, then is more than 1, at this time has a downward correction effect on , making the importance of the th channel smaller. For example, although the contrast of the th channel is relatively high, but due to is also very large, that is, the features are too complex, which is not conducive to distinguishing the vegetation area from other ground objects. By downward correction, the importance of this channel is reduced, which is also in line with the previous logic.

[0089] In summary, through this formula, the importance of each channel in vegetation detection can be accurately evaluated. In practical applications, different channels have different reflection abilities for vegetation and other ground objects. This formula can comprehensively consider factors such as contrast and feature complexity, and assign reasonable importance to each channel.

[0090] After determining the importance of each channel, in the subsequent process of updating remote sensing images, different weights can be assigned according to the importance of each channel, so that the fused image can more prominently display vegetation information and improve the accuracy of vegetation detection.

[0091] S4: Extract the feature vectors of the remote sensing image and its images of each channel.

[0092] A remote sensing image is a complex geospatial data that contains rich ground object information. The data contained in a remote sensing image is high-dimensional and complex. By extracting feature vectors, dimensionality reduction of the data can be achieved, and the feature vectors provide a quantifiable way to measure the similarity or difference between images. By calculating the distance or similarity between feature vectors, it can be determined whether different images belong to the same category, or different ground object types in the remote sensing image can be identified.

[0093] In one embodiment, the feature vectors are extracted according to the following method:

[0094] For a remote sensing image, a histogram of the pixel values of the remote sensing image is constructed. The abscissa of the histogram is different pixel values, and the ordinate is the number of pixel points corresponding to each pixel value. The number of pixel points corresponding to all pixel values in the remote sensing image constitutes the feature vector of the remote sensing image. For example, there are 10 pixel points with a pixel value of 0, 11 pixel points with a pixel value of 1, 12 pixel points with a pixel value of 3, and 110 pixel points with a pixel value of 255 in the remote sensing image (this is a simplified writing, and in an actual remote sensing image, there will not be only these 4 pixel values). The feature vector of the remote sensing image is .

[0095] For each channel image of the remote sensing image, the method of extracting the feature vector is the same as that of the remote sensing image. Taking the red channel image as an example: A histogram of the pixel values of the red channel image is constructed. The abscissa of the histogram is different pixel values, and the ordinate is the number of pixel points corresponding to each pixel value. The number of pixel points corresponding to all pixel values in the red channel image constitutes the feature vector of the red channel image.

[0096] S5: Determine the weights of each channel of the remote sensing image.

[0097] In one embodiment, the weights of each channel of the remote sensing image are calculated based on the following formula:

[0098]

[0099] In this formula, is the weight of the th channel of the remote sensing image, is the cosine similarity between the feature vector of the remote sensing image and the feature vector of the th channel image, is the natural exponential function, is the total number of channels, , is the importance degree of the th channel.

[0100] In this formula, The larger is, the more beneficial the th channel of the remote sensing image is to the segmentation operation of the vegetation area in the remote sensing image, and the weight of the th channel is larger. The smaller is, the more unfavorable the

[0101] In this formula, The size of The difference in pixel value distribution of the channel images, specifically:

[0102] The larger it is, the more similar the -th channel image is to the remote sensing image in pixel value distribution. For image segmentation tasks, especially for the segmentation of vegetation areas, if such a channel image similar to the remote sensing image itself in pixel value distribution is used for segmentation operations, the effect is likely to be similar to directly segmenting the grayscale image of the remote sensing image. For identifying and segmenting vegetation areas, this channel image does not bring additional and unique information or features, and this channel image does not have unique advantages. Therefore, in this case, according to the logic of the formula, the weight of this channel will be smaller, which is a reasonable setting because if the information provided by a channel image is not much different from that of the remote sensing image itself, then its role in complex vegetation area segmentation tasks is relatively limited, and giving this channel a lower weight can avoid this channel having too much influence on the final segmentation result.

[0103] On the contrary, when has a small value, this indicates that the -th channel image is very different from the remote sensing image in pixel value distribution, which implies that this channel image may contain information different from the remote sensing image, and this information may be unique and valuable. For the segmentation task of vegetation areas, when using this channel with a large difference from the remote sensing image for segmentation, different segmentation effects (compared with directly segmenting the remote sensing image) will be produced, and it is more likely to highlight the unique features of the vegetation area in this channel, which are not obvious in the remote sensing image but can be clearly shown in this channel, and this is very helpful for distinguishing vegetation areas from other ground objects. Therefore, such a channel has more potential value for the segmentation of vegetation areas, so according to the logic of this formula, a larger weight will be assigned to this channel to make it play a more important role in subsequent image analysis and processing, helping to more accurately identify and segment the vegetation area.

[0104] In this formula, the part, the natural exponential function plays a role of adjustment and transformation here. When is larger, is a negative number with a large absolute value, making approach 0, which means that for channels similar to the remote sensing image ( is larger), its contribution in weight calculation will be greatly suppressed, which is in line with the logic that when is larger, the importance of the -th channel is lower and a smaller weight is set. When is smaller, The absolute value of is small, making The value of is relatively large, making the The contribution of each channel in the weight calculation is amplified, which is consistent with When smaller, The higher the importance of a channel, the greater the logic of setting the weight.

[0105] In this formula, Part of it is a comprehensive consideration of the The importance of each channel and the influence of the feature similarity between the channel and the remote sensing image make the final weight calculation more comprehensive and accurate. The larger the size) and the more unique the information ( The smaller the value of this part, the larger the value of this part will be, while for channels that are neither important ( smaller) and lack of unique information ( The larger the channel, the smaller the value of this part will be.

[0106] The denominator of this formula plays a normalization role, ensuring that when allocating weights, the weights between different channels are relative, so that the weight allocation of different channels is carried out under overall consideration, and the weight will not be too large or too small to destroy the overall balance, making the weight allocation more reasonable.

[0107] In summary, this formula can reasonably assign different weights to each channel according to the characteristics of each channel and the potential contribution of each channel to vegetation region segmentation, providing a scientific basis for the subsequent use of multi-channel information for vegetation region segmentation tasks of remote sensing images. By giving greater weights to channels that are more conducive to vegetation region segmentation and giving smaller weights to channels that have little effect on vegetation region segmentation, multi-channel information can be used more effectively and the segmentation effect can be optimized.

[0108] S6: Update the pixel value of each pixel in the remote sensing image according to the weight of each channel to obtain an updated remote sensing image.

[0109] The feature differences of different objects in the updated remote sensing image will be more obvious. The channel information related to vegetation will be highlighted by the size of the weight, making the vegetation area more prominent in the updated image, which is conducive to subsequent accurate segmentation.

[0110] In one embodiment, a method of updating a remote sensing image is:

[0111] The pixel values ​​of the pixels in each channel are directly weighted and summed according to the weights of each channel to update the pixel values ​​of the pixels. Specifically, for any pixel in the remote sensing image, the pixel value is updated according to the following weighted summation formula:

[0112]

[0113] In this formula, is the updated pixel value of this pixel point, is the pixel value of this pixel point in the th channel, is the weight of the th channel of the remote sensing image.

[0114] This operation can integrate the information of each pixel point scattered in different channels, so as to obtain a new pixel value that comprehensively reflects the characteristics of this pixel point, which helps to explore the potential connection between the information of different channels and more comprehensively represent the ground object characteristics represented by each pixel point in the image.

[0115] In one embodiment, another method for updating the remote sensing image is as follows: First, if the weight of a certain channel is the smallest, set the weight of this channel to 0; then, perform weighted summation on the pixel values of the pixel point in each channel according to the weights of each channel to realize the update of the pixel value of the pixel point.

[0116] Specifically, for any pixel point in the remote sensing image, its pixel value is updated according to the following formula:

[0117]

[0118] In this formula, is the updated pixel value of this pixel point, is the pixel value of this pixel point in the th channel, is the weight of the th channel of the remote sensing image. There will be a weight of 0 in this formula, which is equivalent to not considering the information of that least important channel when updating the pixel value.

[0119] This operation is because the channel with the smallest weight mainly reflects the ground object details unrelated to vegetation. By setting the weight of this channel to 0, it is possible to avoid the interference of these irrelevant ground object details on the extraction of vegetation characteristics and make the subsequent analysis more focused on the key information.

[0120] S7: Accurately segment the vegetation area from the updated remote sensing image to complete vegetation detection.

[0121] In one embodiment, the classic image binarization algorithm OTSU is used to perform segmentation operation on the updated remote sensing image. The core principle of the OTSU algorithm is to automatically search for an optimal threshold by calculating the gray histogram of the image based on the principle of maximizing the between-class variance, and divide the pixel points in the image into two categories, so that the variance between these two categories reaches the maximum, thereby realizing the effective segmentation of the image.

[0122] In the present invention, after segmenting the updated remote sensing image using the OTSU algorithm, the segmentation result is as follows Figure 9 shown. It only contains two pixel values, 0 and 1. 0 represents white (bright), and 1 represents black (dark). The pixel points with a pixel value of 0 in the segmentation result are determined as vegetation pixel points, and the continuous area composed of all vegetation pixel points is the vegetation area. In this way, the vegetation area can be clearly and accurately extracted from the complex remote sensing image of the mining area, and the vegetation detection of the mining area is completed through these operations.

[0123] In one embodiment, the vegetation detection further includes:

[0124] After extracting the vegetation area, the vegetation coverage rate of the mining area is also calculated. Specifically, the ratio of the number of pixel points in the vegetation area to the number of all pixel points in the updated remote sensing image is used as the vegetation coverage rate of the mining area. This calculation result is of great value for the detection and analysis of the ecological environment of the mining area. A higher vegetation coverage rate often means a relatively good ecological environment in the mining area, and vegetation plays a positive role in soil and water conservation, air purification, climate regulation, etc.; on the contrary, a lower vegetation coverage rate may imply potential risks such as ecological degradation and soil erosion in the mining area. By calculating and detecting the vegetation coverage rate in a long-term and continuous manner, powerful data support can be provided for the ecological restoration, environmental protection, and sustainable development planning of the mining area.

[0125] In the description of this specification, the meanings of "a plurality of" and "several" are at least two, such as two, three, or more, etc., unless otherwise clearly and specifically defined.

[0126] Although this specification has shown and described multiple embodiments of the present invention, it is obvious to those skilled in the art that such embodiments are provided only by way of example. Those skilled in the art will think of many ways of modification, change, and substitution without departing from the idea and spirit of the present invention. It should be understood that various alternative solutions to the embodiments of the present invention described herein can be adopted in the process of practicing the present invention.

Claims

1. A mining area vegetation detection method based on satellite remote sensing images, characterized in that: include: Collect remote sensing images of the mining area and evaluate the feature complexity of the remote sensing images and their channel images: For any image in the remote sensing image and its channel images, the local binary pattern is used to obtain the eigenvalue of each pixel in the image, and a histogram of the eigenvalues ​​is constructed. The horizontal axis of the histogram is different eigenvalues, and the vertical axis is the number of pixels corresponding to each eigenvalue. Calculate the feature complexity of any image in the remote sensing image and its channel images: ;in, is the feature complexity of any image in the remote sensing image and its channel images, is the variance of the number of pixels corresponding to all eigenvalues ​​in the histogram, is the maximum number of pixels corresponding to all eigenvalues ​​in the histogram. is the number of eigenvalues ​​that appear in the image, is the total number of eigenvalues; According to the feature complexity of the remote sensing image and its channel images, as well as the contrast of each channel image, the importance of each channel is calculated ,in, For the The importance of each channel, is the feature complexity of the remote sensing image, For remote sensing images The feature complexity of the channel image, For remote sensing images The contrast of the channel image, is a linear normalization function; Extract the feature vector of the remote sensing image and its channel images: For any image in the remote sensing image and its channel images, construct a histogram of the pixel values ​​of the image. The horizontal axis of the histogram is different pixel values, and the vertical axis is the number of pixel points corresponding to each pixel value. The number of pixel points corresponding to all pixel values ​​in the image constitutes the feature vector of the image; and calculate the weight of each channel of the remote sensing image in combination with the importance of each channel: ,in, For remote sensing images The weight of each channel, is the feature vector of the remote sensing image and The cosine similarity between the feature vectors of the channel images reflects the difference between the remote sensing image and the The difference in pixel value distribution between the channel images is is the natural exponential function, is the total number of channels; For the pixel points in the remote sensing image, the pixel values ​​of the pixel points are updated according to the weights of each channel to obtain an updated remote sensing image. The updated remote sensing image is segmented by the OTSU algorithm, and the vegetation area is determined according to the segmentation results to realize vegetation detection in the mining area.

2. The mining area vegetation detection method based on satellite remote sensing images according to claim 1 is characterized in that: One method to update the pixel value of a pixel point according to the weight of each channel is: The pixel values ​​of the pixel points in each channel are directly weighted and summed according to the weights of each channel to update the pixel values ​​of the pixel points.

3. The mining area vegetation detection method based on satellite remote sensing images according to claim 1 is characterized in that: Another way to update the pixel value according to the weight of each channel is: First, if the weight of a channel is the smallest, set the weight of the channel to 0; Then, the pixel values ​​of the pixel points in each channel are weighted and summed according to the weights of each channel to update the pixel values ​​of the pixel points.

4. The mining area vegetation detection method based on satellite remote sensing images according to claim 1 is characterized in that: The method to determine the vegetation area based on the segmentation results is: The pixel points with a pixel value of 0 in the segmentation result of the updated remote sensing image are regarded as vegetation pixels, and the area formed by all vegetation pixels is regarded as the vegetation area.

5. The mining area vegetation detection method based on satellite remote sensing images according to claim 1 is characterized in that: The contrast of each channel image is any one of Weber contrast, Michelson contrast and root mean square contrast.

6. The mining area vegetation detection method based on satellite remote sensing images according to claim 1 is characterized in that: The channels are red channel, green channel and blue channel.

7. The mining area vegetation detection method based on satellite remote sensing images according to claim 4 is characterized in that: The vegetation detection further comprises: After the vegetation area is determined, the ratio of the number of pixels in the vegetation area to the number of all pixels in the updated remote sensing image is taken as the vegetation coverage rate of the mining area.

Images