A remote sensing image data acquisition method for farmland soil type classification

Abstract

The application relates to the field of remote sensing image processing, in particular to a remote sensing image data acquisition method for farmland soil type classification, which comprises the following steps: periodically acquiring remote sensing image data of a target region according to a preset period; obtaining distance measurement between any two pixel points in each remote sensing image, segmenting each remote sensing image to obtain a ground object region; analyzing the trend and amplitude of the spectrum curve of each pixel point to determine the red edge performance degree of each pixel point; determining the vegetation growth performance degree of each pixel point; determining the farmland soil segmentation quality of each remote sensing image; screening a to-be-classified image from all remote sensing images of the target region, screening a farmland soil region to construct a bare soil mask image, and combining a neural network to perform farmland soil classification on the ground object region in the to-be-classified image. The application aims to accurately divide the farmland soil region in the remote sensing image and improve the accuracy of farmland soil type classification.

Description

Technical Field

[0001] This application relates to the field of remote sensing image processing, specifically to a remote sensing image data acquisition method for farmland soil type classification. Background Technology

[0002] Soil type classification is a crucial step in precision agriculture management. It helps improve crop yields, optimize fertilization and irrigation programs, and promote the sustainable use of arable land resources. Different soil types exhibit significant differences in structure, nutrient content, and water retention capacity, directly impacting crop growth environments and agricultural input methods. Scientific soil type classification enables site-specific planting plans, reduces resource waste, enhances agricultural resilience against disasters, and provides reliable data support for agricultural policy-making, land consolidation, and ecological protection.

[0003] Currently, the classification of farmland soil types using remote sensing images mainly relies on multispectral or hyperspectral remote sensing data. By combining the spectral reflectance characteristics of each pixel in the image at different bands, the spectral feature information of farmland soil is extracted. Machine learning or deep neural network models are then used to classify and identify different types of farmland soil. However, since remote sensing images usually contain a large amount of non-farmland soil information, such as vegetation cover, water reflection, and artificial facilities, these impurity areas have significant spectral differences from real farmland soil. This can interfere with the model's learning of the spectral features of farmland soil, leading to increased training errors, deviations in feature extraction, and blurred classification boundaries, thus seriously affecting the accuracy and stability of classification. Summary of the Invention

[0004] In view of the above, it is necessary to provide a remote sensing image data acquisition method for farmland soil type classification to solve the above problems.

[0005] One embodiment of this application provides a remote sensing image data acquisition method for farmland soil type classification, the method comprising:

[0006] Remote sensing image data of the target area are collected periodically according to a preset cycle;

[0007] The shape fluctuation characteristics of the spectral curves corresponding to any two pixels in each remote sensing image are analyzed to obtain the distance metric between any two pixels in each remote sensing image. Based on the distance metric, each remote sensing image is segmented to obtain several land feature regions.

[0008] The trend and amplitude difference between each maximum and the previous minimum on the spectral curve of each pixel are analyzed to obtain the red edge performance of each maximum on the spectral curve of each pixel; the red edge performance degree of each pixel is determined based on the numerical distribution of the red edge performance of all maximums on the spectral curve of each pixel; and the vegetation growth performance degree of pixels at the same location in all remote sensing images is determined based on the variation characteristics of the red edge performance degree of pixels at the same location in all remote sensing images.

[0009] By combining the distribution of vegetation growth intensity and red edge intensity of all pixels in each land feature region of each remote sensing image, the farmland soil segmentation quality of each remote sensing image is determined. Based on the farmland soil segmentation quality, images to be classified are selected from all remote sensing images of the target area. Based on the distribution of vegetation growth intensity of all pixels in each land feature region of the images to be classified, farmland soil regions are selected to construct bare soil mask images. Combined with the trained neural network, farmland soil is classified in the land feature regions of the images to be classified.

[0010] The process of obtaining the distance metric between any two pixels is as follows:

[0011] Based on the proximity of the band corresponding to the maximum value on the spectral curve of each pixel in each remote sensing image to that of another pixel, obtain the contrast maximum value of each maximum value on the spectral curve of each pixel on the spectral curve of another pixel.

[0012] By analyzing the band difference and amplitude difference between each maximum value on the spectral curve of each pixel and its contrasting maximum value on the spectral curve of another pixel, the distance index between each maximum value on the spectral curve of each pixel and its contrasting maximum value on the spectral curve of another pixel is obtained.

[0013] Calculate the sum of the distance exponents between all maxima on the spectral curve of each pixel and all contrast maxima on the spectral curve of another pixel, denoted as the first sum; calculate the sum of the distance exponents between all maxima on the spectral curve of another pixel and all contrast maxima on the spectral curve of each pixel, denoted as the second sum; take the maximum value of the first sum and the second sum as the distance metric between each pixel and another pixel.

[0014] The process of obtaining the distance index is as follows:

[0015] The band difference and amplitude difference between each maximum value on the spectral curve of each pixel and its contrasting maximum value on the spectral curve of another pixel are obtained and denoted as the first difference and the second difference. The normalized result of the first difference is positively fused with the second difference to obtain the distance index between each maximum value on the spectral curve of each pixel and its contrasting maximum value on the spectral curve of another pixel.

[0016] Specifically, obtaining the contrast maximum value of each maximum value on the spectral curve of each pixel on the spectral curve of another pixel involves:

[0017] For the spectral curve of each pixel in each remote sensing image, extract all the maxima on the spectral curve. For each maximum value of each pixel, find the maximum value of the closest band in the spectral curve of another pixel, and use it as the comparison maximum value of each maximum value on the spectral curve of each pixel to the spectral curve of another pixel.

[0018] Specifically, obtaining the red-edge representation of each maximum value on the spectral curve of each pixel is as follows:

[0019] Calculate the slope of the line connecting each maximum value and its previous minimum value on the spectral curve of each pixel.

[0020] Obtain the amplitude difference between each maximum value and its previous minimum value on the spectral curve of each pixel;

[0021] The normalized slope of the maximum value obtained from the spectral curve of each pixel is positively fused with the amplitude difference, and this result is used as the red edge representation of each maximum value on the spectral curve of each pixel.

[0022] Specifically, determining the degree of red edge representation for each pixel involves:

[0023] Extract the two largest red edges corresponding to the maxima in the spectral curve of each pixel;

[0024] Using the largest red-edge expressiveness as the numerator and the other red-edge expressiveness as the denominator, the normalized value of the resulting fraction is taken as the red-edge expressiveness of each pixel.

[0025] Specifically, determining the vegetation growth performance of pixels at the same location in all remote sensing images involves:

[0026] Obtain the range of red edge representation for pixels at the same location in all remote sensing images; calculate the percentage of the maximum red edge representation for pixels at the same location in all remote sensing images.

[0027] The product of the range value and the numerical percentage is used as the vegetation growth performance of each pixel.

[0028] Specifically, determining the farmland soil segmentation quality for each remote sensing image involves:

[0029] Calculate the average vegetation growth performance of all pixels in each land cover area in each remote sensing image, and record it as the first average.

[0030] Calculate the average red edge representation of all pixels in each land feature area in each remote sensing image, and record it as the second mean;

[0031] The first mean and the second mean of each land feature region in each remote sensing image are positively fused. The sum of all positively fused results for all land feature regions in each remote sensing image is calculated to obtain the farmland soil segmentation quality of each remote sensing image.

[0032] Specifically, the image to be classified is a remote sensing image of farmland soil with the highest segmentation quality.

[0033] Specifically, the process of constructing bare soil mask images by screening farmland soil areas involves:

[0034] In the image to be classified, for any land cover region, the average value of the vegetation growth performance of all pixels is taken as the farmland soil probability of the any land cover region; land cover regions with a farmland soil probability greater than a preset value are recorded as farmland soil regions.

[0035] The non-farmland soil areas in the image to be classified are removed to obtain the bare soil mask image of the image to be classified.

[0036] This application has at least the following beneficial effects:

[0037] In classifying farmland soil types in remote sensing images, this application first divides each remote sensing image into land cover regions based on the differences in pixel spectral characteristics, since vegetation cover affects the classification results. Because crops or weeds in farmland soil images exhibit red edges in summer, this application calculates the degree of red edge representation for each pixel. Due to the vegetation cover on farmland soil, and the cyclical nature of vegetation changes with the seasons, this application uses the variation in red edge representation to determine the vegetation growth performance of each pixel and the probability that it is a farmland soil pixel. Furthermore, by combining the red edge representation and vegetation growth performance of pixels within the land cover region, the image segmentation quality of each remote sensing image is evaluated, thereby achieving accurate division of farmland soil regions and improving the accuracy of farmland soil type classification results. Attached Figure Description

[0038] Figure 1 A flowchart illustrating a remote sensing image data acquisition method for farmland soil type classification provided in this application;

[0039] Figure 2 This is a schematic diagram illustrating the acquisition of the quality of farmland soil segmentation provided in this application. Detailed Implementation

[0040] In the description of the embodiments in this application, the words "exemplary," "or," and "for example" are used to indicate examples, illustrations, or descriptions. Any embodiment or design scheme described as "exemplary" or "for example" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or design schemes. Specifically, the use of the words "exemplary," "or," and "for example" is intended to present the relevant concepts in a specific manner.

[0041] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used in this application's specification is for the purpose of describing particular embodiments only and is not intended to be limiting of the application.

[0042] It should also be noted that the terms "first" and "second" in this application and its accompanying drawings are used to distinguish similar objects, rather than to describe a specific order or sequence. The methods disclosed in the embodiments of this application or the methods shown in the flowcharts include one or more steps for implementing the method. Without departing from the scope of protection of this application, the execution order of multiple steps can be interchanged, and some steps can also be deleted.

[0043] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains.

[0044] This application proposes a remote sensing image data acquisition method for farmland soil type classification, which is applied to the field of remote sensing image processing. (See attached document.) Figure 1 The method includes the following steps:

[0045] S1: Collect remote sensing image data of the target area periodically according to a preset cycle.

[0046] Satellite remote sensing images have the characteristics of wide spatial coverage and strong periodicity, making them suitable for large-scale farmland soil surveys.

[0047] Considering the meteorological conditions of the target area (such as cloud cover and rainfall), images are acquired during sunny and dry periods. Remote sensing images of the target area are obtained using Gaofen-5 satellite remote sensing image data. Each pixel in each remote sensing image contains reflectance information for each band. Starting from the beginning of spring, images are collected periodically according to a preset cycle to obtain one year's worth of remote sensing image data; in this embodiment, the preset cycle is 16 days.

[0048] S2: Analyze the shape fluctuation characteristics of the spectral curves corresponding to any two pixels in each remote sensing image, obtain the distance metric between any two pixels in each remote sensing image, and segment each remote sensing image based on the distance metric to obtain several land feature regions.

[0049] When identifying farmland soil regions in remote sensing images, farmland is characterized by its wide distribution and block-like distribution. Different regions in the image may contain different types of land features. Therefore, it is necessary to first divide the image into blocks based on its content and then divide regions with similar spectral and textural features into the same block based on spatial heterogeneity. This allows for a focus on feature extraction from homogeneous regions in subsequent analysis.

[0050] In the acquired remote sensing images, each pixel corresponds to a spectral vector, which records the reflectance of that pixel in different spectral bands. Different types of ground features exhibit different reflectance distributions across various spectral bands. However, pixels of the same ground feature show similar reflectance distributions across each spectral band.

[0051] The types of ground features considered in this application include: water bodies, farmland soil, vegetation, and artificial structures. Different types of ground feature pixels exhibit different spectral characteristics. Water bodies have low reflectance across all wavelengths. Farmland soils with high organic matter content or dark colors (such as black soil) have reduced visible light reflectance, while sandy or dry farmland soils have higher visible light reflectance. Near-infrared reflectance typically increases, or in some cases, tends to plateau, while short-wave infrared reflectance usually remains at a moderate level. Vegetation has low reflectance in the visible light band because chlorophyll strongly absorbs blue and red light for photosynthesis. In the near-infrared band, the internal cellular structure of plant leaves strongly scatters near-infrared light, leading to increased reflectance. In the short-wave infrared band, reflectance is low because water in leaves strongly absorbs short-wave infrared radiation. Artificial structures, due to the diversity of building materials and colors, may exhibit irregular spectral shapes, depending on the material type and surface properties.

[0052] This application aims to acquire farmland soil regions, thus requiring precise segmentation of these regions. Considering that crops may be grown in these areas, they typically comprise both farmland soil feature pixels and vegetation feature pixels. Both farmland soil feature pixels and vegetation feature pixels exhibit clear spectral variation trends and concentration characteristics; therefore, based on these features, farmland soil feature pixels and vegetation feature pixels can be accurately segmented in remote sensing images.

[0053] This application employs superpixel segmentation to divide remote sensing images, which requires measuring the distance between pixels in the remote sensing image. For any pixel in each remote sensing image, based on the spectral vector of that pixel, the band-reflectance curve of that pixel is obtained and denoted as the spectral curve. All maxima on the spectral curve are extracted. The band corresponding to each maximum value in the spectral curve of that pixel is selected, and the maximum value in the closest band is found in the spectral curves of the remaining pixels. This maximum value is used as the comparative maximum value of each maximum value of that pixel in the spectral curves of the remaining pixels.

[0054] To calculate the distance index between any two pixels in each remote sensing image: First, compare the band difference between each maximum value of any pixel and its contrast maximum value on the spectral curves of the other pixels, denoted as the first difference; calculate the amplitude difference between each maximum value of any pixel and its contrast maximum value on the spectral curves of the other pixels, denoted as the second difference; then, positively fuse the normalized result of the first difference with the second difference to obtain the distance index between each maximum value of any pixel and its contrast maximum value on the other pixels.

[0055] In this embodiment, both the first difference and the second difference are obtained by calculating the absolute value of the difference between variables; the normalization method uses the maximum-minimum normalization method; the positive fusion of multiple variables uses a multiplication calculation method; specifically, the distance exponent between the c-th maximum value of the spectral curve of the a-th pixel in each spectral image and its contrast maximum value on the spectral curve of the b-th pixel is denoted as... Its formula is as follows: In the formula, For the first The spectral curve of the nth pixel A band with a maximum value; For the first The spectral curve of the nth pixel The maximum value at the th ... The band of contrast maxima in the spectral curve of each pixel; For the first The spectral curve of the nth pixel The magnitude of each maximum value; For the first The spectral curve of the nth pixel The maximum value at the th ... The magnitude of the contrast maxima in the spectral curve of each pixel; This represents the normalization function.

[0056] It should be understood that, The larger it is, the more likely it is to be the first The spectral curve of the nth pixel The maximum value and its th value in the th case The greater the difference in the contrast maxima in the spectral curve of each pixel, the greater the difference.

[0057] Further, the sum of the distance indices between all maxima on the spectral curve of any pixel and all contrast maxima on the spectral curves of the remaining pixels is calculated and denoted as the first sum; the sum of the distance indices between all maxima on the spectral curves of the remaining pixels and all contrast maxima on the spectral curve of any pixel is calculated and denoted as the second sum; the maximum value of the first sum and the second sum is used as the distance metric between any pixel and the remaining pixels.

[0058] Based on the distance metric between pixels, a superpixel segmentation algorithm is used to segment each spectral image to obtain several ground feature regions.

[0059] S3: Analyze the trend and amplitude difference between each maximum and the previous minimum on the spectral curve of each pixel to obtain the red edge performance of each maximum on the spectral curve of each pixel; determine the red edge performance degree of each pixel based on the numerical distribution of the red edge performance of all maximums on the spectral curve of each pixel; determine the vegetation growth performance degree of pixels at the same location in all remote sensing images based on the variation characteristics of the red edge performance degree of pixels at the same location in all remote sensing images.

[0060] In remote sensing images, farmland soil areas exhibit distinct temporal variations with the seasons. During spring and summer, exposed farmland soil is often covered by crops or weeds, and its spectral characteristics show typical vegetation features, such as red edges. In autumn and winter, or before planting season, these areas revert to the spectral characteristics of bare soil. Therefore, by analyzing the spectral changes of the same land cover area in seasonal image sequences at different times, it is possible to infer the likelihood that the area belongs to farmland soil.

[0061] Vegetation exhibits a distinct "red edge" characteristic at a wavelength of approximately 700 nm, with reflectance rising sharply at the boundary between the red and near-infrared bands. This is because plant chlorophyll strongly absorbs light energy in the red band for photosynthesis, while it absorbs almost no light in the near-infrared band. The cellular structure inside the leaves results in significant reflection, thus this characteristic can be used to determine the intensity of the red edge in each land cover area.

[0062] Specifically, for each pixel in each feature area, the slope of the line connecting each maximum value to its previous minimum value on the spectral curve of each pixel is calculated; the amplitude difference between each maximum value and its previous minimum value on the spectral curve of each pixel is obtained; the normalized value of the slope obtained from each maximum value on the spectral curve of each pixel is forward-fused with the amplitude difference, and this result is used as the red-edge representation of each maximum value on the spectral curve of each pixel. In this embodiment, the amplitude difference is calculated by the difference between amplitudes; the forward fusion of multiple variables is performed using a multiplication method.

[0063] It should be understood that the greater the slope of the line connecting each maximum value to its previous minimum value, the greater the difference in amplitude, indicating that the spectral curve of the pixel has a greater upward trend at that maximum value.

[0064] In the spectral curve of a single pixel, the true red-edge rise trend is usually significantly unique, that is, there is a prominent reflectance jump near the red to near-infrared band, while no similar steep rise is observed in other bands. Based on the uniqueness of the reflectance jump, the red-edge performance of the pixel can be calculated. Specifically, the two largest red-edge performance values ​​corresponding to the maxima in the spectral curve of each pixel are extracted. The largest red-edge performance value is used as the numerator, and the other red-edge performance value is used as the denominator. The normalized value of the obtained fraction is taken as the red-edge performance of each pixel.

[0065] In farmland soil classification, considering the significant seasonal variations in vegetation growth, the red edge features of the same pixel in remote sensing images acquired at different time periods are affected by the crop growth cycle. Therefore, the red edge features from a single time period are often insufficient to accurately reflect the long-term land cover attributes of that pixel. Thus, it is necessary to combine multi-temporal remote sensing images to analyze the red edge variation trends of the same pixel across different seasons.

[0066] From a temporal perspective, if a pixel belongs to farmland soil, its red edge feature typically exhibits significant seasonal fluctuations: after spring sowing, the red edge feature gradually increases from a low value; in summer, when vegetation is growing vigorously, the red edge feature reaches its peak; in autumn, as the crop matures, the red edge feature gradually decreases; and in winter, if it is during fallow or tilling season, the red edge feature drops to a low value again. In contrast, pixels that are long-term exposed or non-vegetated areas show smaller fluctuations in their red edge feature throughout the year, typically remaining at a low level.

[0067] Based on this, remote sensing images of the same target area at all times are analyzed to obtain the range of red edge performance at the same location in all remote sensing images; the percentage of the maximum red edge performance at the same location in all remote sensing images is calculated; and the product of the range and the percentage is taken as the vegetation growth performance of each pixel. It should be understood that the larger the range, the greater the variation in red edge performance at each pixel across all times, reflecting a greater degree of growth dynamics; the larger the percentage, the more typical red edge phenomenon exists for that pixel in a single season.

[0068] S4: Determine the farmland soil segmentation quality of each remote sensing image by combining the distribution of vegetation growth and red edge performance of all pixels in each land feature region of each image; based on the farmland soil segmentation quality, select images to be classified from all remote sensing images of the target area; based on the distribution of vegetation growth in each land feature region of the images to be classified, construct bare soil mask images of farmland soil regions; and combine the trained neural network to classify farmland soil in the land feature regions of the images to be classified.

[0069] Accurate segmentation of farmland soil areas in remote sensing images is a crucial prerequisite for farmland soil type classification. However, in actual imagery, farmland soil is often obscured by vegetation due to seasonal changes, especially in summer when crop or weed growth prevents direct exposure to remote sensing, affecting classification accuracy. Therefore, it is essential to ensure that the acquired area represents farmland soil without vegetation cover. Consequently, the quality of farmland soil segmentation in each remote sensing image needs to be judged based on the degree of red border representation of the farmland soil area.

[0070] Based on this, the farmland soil segmentation quality of each remote sensing image is calculated. Specifically: the average vegetation growth performance of all pixels in each land cover region of each remote sensing image is calculated and denoted as the first average; the average red edge performance of all pixels in each land cover region of each remote sensing image is calculated and denoted as the second average; the first average and the second average of each land cover region in each remote sensing image are forward fused, and the sum of all forward fusion results obtained for all land cover regions in each remote sensing image is calculated to obtain the farmland soil segmentation quality of each remote sensing image. In this embodiment, the forward fusion between variables is performed using a multiplication calculation method.

[0071] The schematic diagram for obtaining the quality of farmland soil segmentation is shown below. Figure 2 As shown.

[0072] It should be understood that the greater the average vegetation growth performance, the more likely that each land feature area in each remote sensing image is a farmland soil area, and the better the area is delineated as farmland soil. At the same time, the greater the average red edge performance, the more likely that the land feature area is not only a farmland soil area, but also that the farmland soil in the area is not covered by vegetation, which is beneficial for the subsequent identification of farmland soil types.

[0073] The remote sensing image with the highest quality of farmland soil segmentation is recorded as the image to be classified. In the image to be classified, for any land feature area, the average value of the vegetation growth performance of all pixels is taken as the farmland soil probability of that land feature area. Land feature areas with a farmland soil probability greater than a preset value are recorded as farmland soil areas. In this embodiment, the preset value is 0.7, and the implementer can adjust the size of the preset value according to the actual situation.

[0074] In the image to be classified, a bare soil mask map is first constructed by removing non-farmland soil areas, thus obtaining the farmland soil image to be classified. Based on this method, farmland soil images of different regions are generated as a sample set. Subsequently, the locations of farmland soil areas in the remote sensing images are manually labeled and classified, with labels including but not limited to: red soil, black soil, saline soil, and sandy soil. The dataset will be divided into 70% for training, 15% for validation, and 15% for testing.

[0075] During training, this application uses the cross-entropy loss function as the objective function. After the model training is complete, a new image to be classified is input, and the farmland soil type corresponding to each land cover area is output. The neural network model used in this application is a convolutional neural network model, which is a well-known existing technology and will not be described in detail here.

[0076] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. In some alternative implementations, the functions marked in the blocks may occur in a different order than that shown in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. In the descriptions corresponding to the flowcharts and block diagrams in the accompanying drawings, the operations or steps corresponding to different blocks may also occur in a different order than disclosed in the description; sometimes there is no specific order between different operations or steps. For example, two consecutive operations or steps may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. Each block in a block diagram and / or flowchart, and combinations of blocks in a block diagram and / or flowchart, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0077] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A remote sensing image data collection method for farmland soil type classification, characterized by, The method comprises the following steps: Periodically collecting remote sensing image data of the target area according to a preset period; Analyzing the shape fluctuation characteristics of the spectral curves corresponding to any two pixel points in each remote sensing image, obtaining the distance measure between the any two pixel points in each remote sensing image, segmenting each remote sensing image based on the distance measure, and obtaining a plurality of ground object regions; Analyzing the trend degree and amplitude difference between each maximum value on the spectral curve of each pixel point and the previous minimum value, obtaining the red edge performance of each maximum value on the spectral curve of each pixel point, determining the red edge performance degree of each pixel point according to the numerical distribution of the red edge performance of all maximum values on the spectral curve of each pixel point, and determining the vegetation growth performance degree of the pixel points at the same position in all remote sensing images according to the change characteristics of the red edge performance degree of the pixel points at the same position in all remote sensing images; Determine the farmland soil segmentation quality of each remote sensing image by comprehensively considering the distribution of the vegetation growth performance degree and the distribution of the red edge performance degree of all pixel points in each ground object region of each remote sensing image, select the image to be classified from all remote sensing images of the target area based on the farmland soil segmentation quality, select the farmland soil region to construct the bare soil mask image based on the distribution of the vegetation growth performance degree of all pixel points in each ground object region in the image to be classified, and combine the trained neural network to classify the ground object region in the image to be classified. The distance measure between the any two pixel points is obtained by: According to the proximity of the maximum value on the spectral curve of each pixel point to the other pixel point, the comparative maximum value of each maximum value on the spectral curve of each pixel point on the spectral curve of the other pixel point is obtained. The distance index between each maximum value on the spectral curve of each pixel point and its comparative maximum value on the spectral curve of the other pixel point is obtained by analyzing the band difference and amplitude difference between each maximum value on the spectral curve of each pixel point and its comparative maximum value on the spectral curve of the other pixel point. The accumulation sum of the distance index between all maximum values on the spectral curve of each pixel point and all comparative maximum values on the spectral curve of the other pixel point is calculated, which is denoted as the first accumulation sum. The accumulation sum of the distance index between all maximum values on the spectral curve of the other pixel point and all comparative maximum values on the spectral curve of each pixel point is calculated, which is denoted as the second accumulation sum. The maximum value of the first accumulation sum and the second accumulation sum is taken as the distance measure between each pixel point and the other pixel point.

2. The remote sensing image data collection method for the farmland soil type classification according to claim 1, wherein, The distance index is obtained by: The band difference and amplitude difference between each maximum value on the spectral curve of each pixel point and its comparative maximum value on the spectral curve of the other pixel point are obtained, which are denoted as the first difference and the second difference. The normalized result of the first difference is fused with the second difference in a positive direction to obtain the distance index between each maximum value on the spectral curve of each pixel point and its comparative maximum value on the spectral curve of the other pixel point.

3. The method of claim 1, wherein the method is a method of collecting remote sensing image data for farmland soil type classification, characterized by, The acquisition of each pixel point spectrum curve on each maximum value in another pixel point spectrum curve contrast maximum value, specifically for: For each pixel point spectrum curve in each remote sensing image, extract all the maximum values on the spectrum curve, for each maximum value corresponding to each pixel point, find the closest maximum value on the spectrum curve of another pixel point, and take it as the contrast maximum value of each maximum value on the spectrum curve of another pixel point spectrum curve.

4. The method of claim 1, wherein the method is a method of collecting remote sensing image data for farmland soil type classification, characterized by, The red edge performance of each maximum value on the spectrum curve of each pixel point is obtained, specifically: Calculate the slope of the line connecting each maximum value on the spectrum curve of each pixel point and its previous minimum value; Obtain the amplitude difference between each maximum value on the spectrum curve of each pixel point and its previous minimum value; The normalized value of the slope of each maximum value on the spectrum curve of each pixel point is fused with the amplitude difference, and the result is taken as the red edge performance of each maximum value on the spectrum curve of each pixel point.

5. The method for collecting remote sensing image data for the classification of farmland soil types according to claim 1, wherein, The red edge performance of each pixel point is determined, specifically: Extract the maximum two red edge performances corresponding to the maximum value in the spectrum curve of each pixel point; Take the maximum red edge performance as the numerator and the other red edge performance as the denominator, and take the normalized value of the obtained score as the red edge performance of each pixel point.

6. The remote sensing image data collection method for farmland soil type classification according to claim 1, wherein, The vegetation growth performance of the same position pixel points in all remote sensing images is determined, specifically: Obtain the range value of the red edge performance of the same position pixel points in all remote sensing images; calculate the numerical value of the maximum value of the red edge performance of the same position pixel points in all remote sensing images; The product between the range value and the numerical value is taken as the vegetation growth performance of each pixel point.

7. The method of claim 1, wherein the method is a method of collecting remote sensing image data for farmland soil type classification, characterized by, The farmland soil segmentation quality of each remote sensing image is determined, specifically: Calculate the mean value of the vegetation growth performance of all pixel points in each ground object region in each remote sensing image, denoted as the first mean value; Calculate the mean value of the red edge performance of all pixel points in each ground object region in each remote sensing image, denoted as the second mean value; Calculate the first mean value and the second mean value of each ground object region in each remote sensing image, calculate the sum of all positive fusion results of all ground object regions in each remote sensing image, and obtain the farmland soil segmentation quality of each remote sensing image.

8. The method for collecting remote sensing image data for the classification of farmland soil types according to claim 1, wherein, The image to be classified is specifically the remote sensing image with the maximum farmland soil segmentation quality.

9. The method for collecting remote sensing image data for the classification of farmland soil types according to claim 1, wherein, The screening of farmland soil region to construct a bare soil mask image, specifically: In the image to be classified, for any ground object region, the mean value of the vegetation growth performance of all pixel points is taken as the farmland soil probability of the ground object region; the ground object region with a farmland soil probability greater than a preset value is recorded as a farmland soil region; The non-farmland soil region in the image to be classified is removed to obtain a bare soil mask image of the image to be classified.

Images