A farmland soil state recognition method based on hyperspectral remote sensing image processing

Abstract

The application discloses a farmland soil state recognition method based on hyperspectral remote sensing image processing and relates to the technical fields of remote sensing image processing and soil state recognition and detection. The method comprises the following steps: acquiring a hyperspectral remote sensing image of a farmland area, determining a ridge direction and converting the ridge direction into a ridge cross-section coordinate image; determining a shadow darkening response and a shadow darkening center line according to a darkening degree and a multi-band spectral shape similarity degree; determining a wet absorption response and a wet absorption center line according to an image space difference of an envelope absorption response; calculating a dislocation stability weight based on a relative dislocation relationship and a dislocation fluctuation degree of the two center lines, and combining the shadow darkening response and the wet absorption response to generate a corrected soil wet state index; and finally performing ordered state division and inverse transformation to output a farmland soil state recognition map. The application can reduce the interference of ridge shadow darkening on soil wet recognition and improve the accuracy of ridge farmland soil state recognition and detection.

Description

Technical Field

[0001] This invention relates to the field of remote sensing image processing and soil condition identification and detection technology, and in particular to a method for identifying farmland soil condition based on hyperspectral remote sensing image processing. Background Technology

[0002] In ridge-cultivated farmland for crops such as corn, potatoes, peanuts, and vegetables, the ridge top, ridge slope, and furrow form a continuous, undulating micro-topographical structure. After rainfall or irrigation, moisture typically migrates along the ridge slope to the furrow or lower part of the ridge slope, creating locally moist areas. With the application of UAV, airborne, and satellite hyperspectral remote sensing imaging technologies in agricultural monitoring, using hyperspectral remote sensing images to detect the moisture status of farmland soil has become an important means of precision irrigation and farmland management. However, in hyperspectral images of ridge-cultivated farmland, shaded ridge slopes, due to insufficient light, form dark bands extending along the ridge direction, while truly moist areas show decreased reflection and enhanced absorption due to increased soil moisture. These two types of areas are easily adjacent in the image and their positions are not entirely consistent, causing the soil condition detection and identification results to shift along the cross-sectional direction of the ridge, making it difficult to accurately distinguish between shaded dark areas on the ridge slope and truly moist areas.

[0003] Existing methods for detecting and identifying farmland soil conditions mostly rely on reflectance intensity, moisture-related band response, or humidity index to directly determine soil moisture levels. They typically use low-reflectance areas in images as a key indicator of humidity, but they fail to adequately consider the brightness differences caused by the sun-shaded sides of ridges in ridged farmland. Especially under imaging conditions with clearly defined ridge orientation and a stable solar incidence direction, the darkening response of shaded ridges may be misinterpreted as increased soil moisture content, while the actual moisture absorption location in the furrows or lower part of the ridge may be misdirected by the darkening zone, causing a shift in the boundary of the humid area. Therefore, existing methods struggle to effectively correct for the misalignment between shaded darkening locations and moisture absorption locations in ridged farmland, affecting the accuracy of farmland soil condition identification maps and their subsequent agricultural management applications. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of existing technologies, such as the easy confusion between shaded and darkened areas of ridge slopes and actual moist absorption areas in hyperspectral remote sensing images of ridged farmland, which leads to a shift in soil dryness and wetness identification results along the cross-sectional direction of the ridge and makes it difficult to accurately output farmland soil condition identification maps. Therefore, this invention proposes a farmland soil condition identification method based on hyperspectral remote sensing image processing.

[0005] To address the problems existing in the prior art, the present invention adopts the following technical solution: A method for identifying farmland soil conditions based on hyperspectral remote sensing image processing includes: S1. Obtain a hyperspectral remote sensing image of the farmland area, determine the ridge direction based on the ridge texture in the hyperspectral remote sensing image, and convert the hyperspectral remote sensing image to a coordinate system with the ridge direction perpendicular to the ridge direction as the cross-sectional direction to obtain a ridge cross-sectional coordinate image. S2. Based on the degree of darkening in the cross-sectional coordinate image of the ridge and the similarity of the multi-band spectral shape, determine the shading darkening response and extract the shading darkening center line on the cross-section. S3. Perform envelope absorption analysis on the cross-sectional coordinate image of the ridge, and determine the wetting absorption response based on the difference in the obtained envelope absorption response in the image space, and extract the wetting absorption center line on the cross-section. S4. Calculate the misalignment stability weight based on the relative misalignment relationship and degree of misalignment fluctuation between the shading and darkening center line and the moisture absorption center line in the cross-sectional direction. S5. Based on the misaligned stability weights, shading darkening response, and wetting absorption response, a corrected soil wetting state index is generated. S6. Based on the corrected soil moisture state index, the soil is divided into ordered states to obtain the classification results under the ridge coordinates. The classification results are then transformed back to the original image coordinate system to obtain the farmland soil state recognition map.

[0006] Preferably, determining the ridge direction based on the ridge texture in the hyperspectral remote sensing image and converting the hyperspectral remote sensing image to a coordinate system with a cross-sectional direction perpendicular to the ridge direction includes: Robust normalization processing is performed on the reflection response of each band in the acquired hyperspectral remote sensing image; Based on the normalized reflection response of each band, the multi-band comprehensive brightness response is calculated, and the ridge texture brightness map is constructed. The brightness map of the ridge texture is sampled or projected along different candidate directions to form a brightness profile of the candidate direction. Calculate the ridge repeat correlation intensity of the candidate direction brightness profile under adjacent displacements, and determine the candidate direction that produces the maximum ridge repeat correlation intensity as the ridge direction; Using the ridge direction as the longitudinal direction and the direction perpendicular to the ridge direction as the cross-sectional direction, coordinate transformation is performed on the normalized hyperspectral remote sensing image to obtain the ridge cross-sectional coordinate image.

[0007] Preferably, the shading darkening response is determined based on the degree of darkening in the cross-sectional coordinate image of the ridge in the cross-sectional direction and the similarity of the multi-band spectral shape, and the shading darkening center line is extracted on the cross-section, including: Calculate the multi-band composite brightness of pixels in the cross-sectional coordinate image of the ridge; The degree of darkening is calculated based on the relative relationship between the multi-band combined brightness of pixels within the same cross-section and the minimum and maximum multi-band combined brightness within that cross-section. The correlation between the normalized spectral shape of the current pixel and the average spectral shape of the current cross-section is calculated to obtain the spectral shape preservation coefficient; The darkening level is fused with the spectral shape preservation coefficient to obtain the shading darkening response; Extract the location of the maximum response of the shading darkening on each cross section, and connect the locations of the maximum response on each cross section to form the shading darkening center line.

[0008] Preferably, envelope absorption analysis is performed on the cross-sectional coordinate image of the ridge, and the wetting absorption response is determined based on the difference in the obtained envelope absorption response in the image space, and the wetting absorption center line is extracted on the cross-section, including: The spectral curves of each pixel in the cross-sectional coordinate image of the ridge are subjected to continuous envelope processing to obtain the envelope reflection response; The envelope absorption response is obtained based on the absorption depth of the cross-sectional coordinate image of the ridge relative to the envelope reflection response. Within the effective band range of soil moisture absorption changes, the band with the greatest spatial difference in envelope absorption response in the current image is identified as the moisture response band. The envelope absorption response corresponding to the wetting response band is taken as the wetting absorption response; The location of the maximum response to the wetting absorption is extracted on each cross-section, and the locations of the maximum response on each cross-section are connected to form the center line of the wetting absorption.

[0009] Preferably, the misalignment stability weight is calculated based on the relative misalignment relationship and degree of misalignment fluctuation between the shading and darkening center line and the moisture absorption center line in the cross-sectional direction, including: Calculate the center misalignment of the moisture absorption centerline relative to the shading and darkening centerline on the same cross-section; Calculate the median value of the center misalignment corresponding to different cross sections, calculate the absolute deviation of each center misalignment relative to the median value of the center misalignment, and use the median value of the absolute deviation as the degree of misalignment fluctuation. The misalignment stability weight is calculated based on the absolute value of the median of the central misalignment and the degree of misalignment fluctuation. Wherein, the larger the absolute value of the median value of the central misalignment and the smaller the degree of misalignment fluctuation, the larger the misalignment stability weight; the smaller the absolute value of the median value of the central misalignment or the larger the degree of misalignment fluctuation, the smaller the misalignment stability weight.

[0010] Preferably, a corrected soil moisture state index is generated based on misalignment stabilization weights, shading darkening response, and moisture absorption response, including: The cross-sectional distribution width around the shading darkening center line is calculated based on the shading darkening response, and a shading effect distribution reflecting the degree to which a pixel approaches the shading darkening center line is constructed based on this cross-sectional distribution width. The cross-sectional distribution width around the wetting absorption centerline is calculated based on the wetting absorption response, and a wetting influence distribution reflecting the degree to which a pixel is close to the wetting absorption centerline is constructed based on this cross-sectional distribution width. The moisture absorption response was robustly normalized to obtain a standardized moisture absorption response. By combining the misalignment stabilization weights, the distribution of shading effects, and the distribution of wetting effects, the standardized wetting absorption response is subjected to shading suppression and wetting reinforcement calculations. This results in the suppression of pixel responses that are close to the shading darkening center line and far from the wetting absorption center line, while the reinforcement of pixel responses that are close to the wetting absorption center line, thus generating a corrected soil wetting state index.

[0011] Preferably, the step of combining misalignment stabilization weights, shading effect distribution, and wetting effect distribution to calculate shading suppression and wetting reinforcement for the standardized wetting absorption response includes: The difference obtained by subtracting the wetting effect distribution from the baseline value of 1, multiplied by the shading effect distribution and the misalignment stability weight, yields the inhibition factor. The result of subtracting the suppression factor from the baseline value of 1 is multiplied by the normalized wetting absorption response of the current pixel to obtain the baseline response retention value. The standardized wetting absorption response at the center line of the wetting absorption cross-section of the current pixel is multiplied by the wetting influence distribution and misalignment stability weights to obtain the center wetting reinforcement value. The soil moisture state index after correction for the current pixel is obtained by adding the base response retention value and the center moisture reinforcement value.

[0012] Preferably, the ordered state is divided according to the corrected soil moisture state index to obtain the classification result in the ridge coordinate system, and the classification result is inversely transformed back to the original image coordinate system, including: Determine the number of soil condition grades; Based on the number of soil condition levels, the corrected soil moisture state index is clustered and classified into ordered states to obtain each soil condition category with a corresponding average moisture state value. Soil condition categories are sorted in ascending order of mean moisture content to obtain soil condition classification results under ridge coordinates. The soil state classification results under the ridge coordinate system are subjected to an inverse transformation from the ridge coordinate system to the original image coordinate system to obtain a farmland soil state identification map.

[0013] Compared with the prior art, the beneficial effects of the present invention are: 1. This invention acquires hyperspectral remote sensing images of farmland areas with ridge texture, determines the ridge direction based on the ridge texture, and transforms the image to the ridge cross-sectional coordinate system, so that the ridge top, ridge slope, and ridge furrow are unfolded in the cross-sectional direction. Furthermore, it extracts the shading darkening response and shading darkening center line based on the degree of darkening within the cross-section and the similarity of multi-band spectral shapes, and extracts the wetting absorption response and wetting absorption center line based on the differences in the envelope absorption response in the image space. Thus, it is possible to locate the shading darkening position and the actual wetting absorption position of the ridge slope in the same coordinate system, avoiding the direct identification of the darkening zone formed by the backlit ridge slope as a wet area, thereby improving the reliability of soil dryness and wetness detection and judgment in hyperspectral images of ridged farmland.

[0014] 2. This invention calculates the misalignment stability weight based on the relative misalignment relationship and degree of misalignment fluctuation between the shading darkening center line and the moisture absorption center line in the cross-sectional direction. It then combines the shading influence distribution, the moisture influence distribution, and the standardized moisture absorption response to generate a corrected soil moisture state index. This suppresses the pixel responses that are close to the shading darkening center line and far from the moisture absorption center line, while preserving and strengthening the pixel responses that are close to the moisture absorption center line. This reduces the traction effect of shading darkening on soil moisture identification, making the final output farmland soil state identification map closer to the actual location distribution of moist areas. It also reduces the probability of the soil dryness and wetness state identification results shifting along the cross-sectional direction of the ridge, and provides more accurate image data for precision irrigation, farmland monitoring, and operation area positioning. Attached Figure Description

[0015] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, illustrate exemplary embodiments of the invention and, together with their description, serve to explain the invention and do not constitute an undue limitation thereof. In the drawings: Figure 1 A flowchart illustrating a method for identifying farmland soil conditions based on hyperspectral remote sensing image processing provided by the present invention; Figure 2 A functional module diagram of a farmland soil condition identification system based on hyperspectral remote sensing image processing provided by the present invention; Figure 3 This is a schematic diagram illustrating the application scenario of ridge farming provided by the present invention. Detailed Implementation

[0016] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0017] Example: This example provides a method for identifying farmland soil conditions based on hyperspectral remote sensing image processing. See [link to example]. Figure 1 Specifically, including: S1. Obtain a hyperspectral remote sensing image of the farmland area, determine the ridge direction based on the ridge texture in the hyperspectral remote sensing image, and convert the hyperspectral remote sensing image to a coordinate system with the ridge direction perpendicular to the ridge direction as the cross-sectional direction to obtain a ridge cross-sectional coordinate image. S2. Based on the degree of darkening in the cross-sectional coordinate image of the ridge and the similarity of the multi-band spectral shape, determine the shading darkening response and extract the shading darkening center line on the cross-section. S3. Perform envelope absorption analysis on the cross-sectional coordinate image of the ridge, and determine the wetting absorption response based on the difference in the obtained envelope absorption response in the image space, and extract the wetting absorption center line on the cross-section. S4. Calculate the misalignment stability weight based on the relative misalignment relationship and degree of misalignment fluctuation between the shading and darkening center line and the moisture absorption center line in the cross-sectional direction. S5. Based on the misaligned stability weights, shading darkening response, and wetting absorption response, a corrected soil wetting state index is generated. S6. Based on the corrected soil moisture state index, the soil is divided into ordered states to obtain the classification results under the ridge coordinates. The classification results are then transformed back to the original image coordinate system to obtain the farmland soil state recognition map.

[0018] In one embodiment of the present invention, a hyperspectral remote sensing image of a farmland area is acquired, the ridge direction is determined based on the ridge texture in the hyperspectral remote sensing image, and the hyperspectral remote sensing image is converted to a coordinate system with the ridge direction perpendicular to the ridge direction as the cross-sectional direction to obtain a ridge cross-sectional coordinate image, including: In detail, this embodiment provides a method for identifying farmland soil conditions based on hyperspectral remote sensing image processing. It is applicable to soil dryness and wetness identification and detection scenarios in ridged farmland areas with ridge texture, such as corn, potatoes, peanuts, and vegetables. Hyperspectral remote sensing images of farmland areas with ridge texture can be acquired through UAV hyperspectral cameras, airborne hyperspectral sensors, or satellite hyperspectral sensors. The hyperspectral remote sensing image is a data cube containing spatial location and reflectance responses of multiple spectral bands. After acquisition, the hyperspectral remote sensing image can be radiometrically corrected, geometrically corrected, cropped for farmland areas, and invalid pixels removed to ensure that each pixel in the input image corresponds to a valid soil or ridge surface area within the farmland area. Suppose the acquired hyperspectral remote sensing image is:

[0019] in, This indicates the pixel location in the original hyperspectral remote sensing image. , band The reflection response value at that time; This represents the horizontal pixel coordinates in the original image coordinate system; Represents the vertical pixel coordinates in the original image coordinate system; The spectral bands in the hyperspectral image are represented. The above three-dimensional data format is adopted because this embodiment needs to use the ridge texture in the image space to determine the ridge direction, and also needs to use the reflection response of different bands to distinguish between shading and darkening and true moisture absorption. A single grayscale image or a single band image is difficult to express the ridge structure and water absorption information at the same time. Robust normalization of the reflection responses in each band yields normalized hyperspectral images:

[0020] in, This represents the normalized hyperspectral reflectance response; This represents the original reflection response before normalization; Indicates band Median reflectance response within the effective pixels of the farmland area; Indicates band Median absolute deviation within the effective pixels of the farmland area; This represents the smallest positive number used to avoid division by zero. The robust normalization using median reflectance response and median absolute deviation is because there may be local strong reflections, shadows, vegetation residues or sensor noise points in farmland images. The median statistic is not sensitive to anomalous pixels, which can bring the reflectance responses of different bands into a comparable scale, while not amplifying local anomalous bright and dark points into ridge structure information. Construct a mound texture brightness map based on the normalized hyperspectral image:

[0021] in, Represents the pixel position in the original image coordinate system Brightness response of the texture at the location; This represents the normalized hyperspectral reflectance response; This indicates that the median value is taken along the spectral band direction. The purpose of constructing the ridge texture brightness map using multi-band median values ​​is to obtain an overall brightness expression that is not sensitive to a single water-sensitive band or a single noise band, so that the periodic strip texture formed by the ridge top, ridge slope and ridge furrow can be stably displayed in the image space. The ridge texture brightness map is sampled or projected along different candidate directions to form a candidate direction brightness profile, and the ridge repetition correlation intensity of the candidate direction brightness profile under adjacent displacements is calculated; for any candidate direction The ridge texture brightness map can be unfolded into a brightness profile along this direction, and the corresponding repetitive correlation intensity can be calculated:

[0022]

[0023] in, Indicate candidate direction The correlation strength of the repeating structure under the ridge; Indicate candidate directions; Indicates the brightness map of the ridge texture along the candidate direction After sampling or projection, at the profile position The brightness profile value formed at that location; Indicates the sampling position on the candidate direction brightness profile; Indicate candidate direction The adjacent displacements are adaptively determined by the repeating texture of the ridge body; This indicates a correlation calculation; The direction that maximizes the repeat correlation intensity of the ridge is used as the identified ridge direction. The repeat correlation intensity is used to determine the ridge direction because ridged farmland has a continuous strip structure in the ridge direction. When the brightness profile is unfolded along the ridge direction, the adjacent repeat relationship is more stable. Therefore, the maximum repeat correlation intensity can reflect the main direction of the ridge texture. Adjacent displacement The adjacent displacements are adaptively determined based on the repeating texture of the ridges in the brightness profile of the candidate direction. Specifically, for any candidate direction brightness profile, the autocorrelation curve of the brightness profile is first calculated along the candidate direction. After excluding the autocorrelation peaks corresponding to zero displacement, the displacements corresponding to stable local peaks that can reflect the repeating texture of the ridges are selected as the adjacent displacements under that candidate direction. When there are multiple local peaks under the same candidate direction, the displacements corresponding to the local peaks that are highly consistent with the statistical results of the furrow spacing or the calibration information of the crop row spacing are selected first. When the statistical results of the furrow spacing or the calibration information of the crop row spacing are not available, the displacements corresponding to the local peaks that change continuously between adjacent candidate directions and have a high corresponding correlation intensity are selected. In this way, the adjacent displacements are determined by the repeating texture relationship formed by the ridge top, ridge slope and furrow in the image, which can adapt to hyperspectral images of ridged farmland under different crop row spacings, different spatial resolutions and different imaging heights.

[0024] After determining the ridge direction, taking the ridge direction as the longitudinal direction and the direction perpendicular to the ridge direction as the cross-sectional direction, a coordinate transformation is performed on the normalized hyperspectral image to obtain the ridge cross-sectional coordinate image:

[0025] in, This represents a hyperspectral image in the coordinate system of the cross-section of the ridge. Indicates the directional coordinates of the cross-section of the ridge; Indicates the coordinates along the ridge direction; Indicates spectral band; Indicates based on the direction of the ridge Established coordinate transformation operation; This represents the normalized hyperspectral image in the original image coordinate system. This coordinate transformation can be achieved using rotation, affine transformation, or resampling transformation based on the ridge direction, ensuring that the ridge tops, ridge slopes, and furrows in the original image are aligned in the cross-sectional direction. Expand upwards and along the ridge direction Maintain continuity.

[0026] It should be noted that the ridge cross-sectional coordinate image in this embodiment refers to image data formed by rearranging the coordinates of a hyperspectral remote sensing image of a farmland area with ridge texture according to the identified ridge direction. In this image data, the ridge direction is used as the longitudinal coordinate, and the direction perpendicular to the ridge direction is used as the cross-sectional coordinate, so that the ridge top, ridge slope, and ridge furrow in the original image are unfolded in the cross-sectional direction, and the same ridge is kept continuous along the ridge direction. Through this image data, the shading and darkening positions of the ridge slope and the soil moisture absorption positions can be compared within the same cross-section, providing a unified coordinate basis for subsequent calculation of center misalignment. The output ridge cross-sectional coordinate image serves as the common input for subsequent shading and darkening response extraction and moisture absorption response extraction, thereby ensuring that the shading center and the moisture center are calculated under the same ridge cross-sectional coordinate system, avoiding the inability to compare misalignment due to inconsistencies in coordinate systems.

[0027] In one embodiment of the present invention, the shading darkening response is determined based on the degree of darkening in the cross-sectional coordinate image of the ridge in the cross-sectional direction and the similarity of the multi-band spectral shape, and the shading darkening center line is extracted on the cross-section, including: Specifically, the multi-band composite brightness of pixels is calculated in the cross-sectional coordinate image of the ridge:

[0028] in, Represents the pixel position in the cross-sectional coordinate system of the ridge. Multi-band integrated brightness at the location; This indicates the position of the pixel in the cross-sectional coordinate image of the ridge in the band. The reflection response at the location; This indicates that the median value is taken along the spectral band direction; the multi-band comprehensive brightness is used to reflect the overall decrease in multi-band reflectance caused by shading, rather than relying solely on the local absorption changes of a certain moisture-sensitive band. For the same location along the ridge direction For a corresponding cross-section, calculate the degree of darkening of the pixels within that cross-section:

[0029] in, Indicates pixel position The degree of darkness in the area; This indicates the combined brightness of the pixel across multiple bands. Indicates the position along the ridge direction Minimum multi-band combined brightness within the cross-section; Indicates the position along the ridge direction The maximum multi-band combined brightness within the cross-section; This indicates a very small positive number used to avoid division by zero. The degree of darkening is calculated by using the relative brightness relationship within the cross-section because different cross-sections may be affected by changes in sensor viewing angle, solar altitude angle and surface albedo. Directly using absolute brightness can easily lead to incomparability between different cross-sections. By normalizing the maximum and minimum brightness within the same cross-section, the backlight darkening position on that cross-section can be highlighted.

[0030] To further distinguish between the overall darkening caused by ridge shading and the enhanced water absorption caused by actual soil moisture, the correlation between the normalized spectral shape of the current pixel and the average spectral shape of the current cross-section is calculated to obtain the spectral shape preservation coefficient; first, the average spectrum of the current cross-section is calculated:

[0031] in, Indicates the position along the ridge direction The cross section at the band The average spectral response under these conditions; This indicates the number of valid pixels within the cross-section; Indicates the position along the ridge direction The set of effective pixels within the cross-section; Represents the pixel position in the cross-sectional coordinate image of the ridge. Spectral response at; Calculate the spectral shape preservation factor:

[0032] in, Indicates pixel position Spectral shape preservation coefficient at that location; Indicates pixel position The frequency bands are divided into each band. The spectral vector formed; Indicates the position along the ridge direction The average spectral vector of the cross section at that location; Represents the L2 norm; This indicates a correlation calculation; This represents a very small positive number used to avoid the vector norm being zero. When the correlation calculation result is less than zero, the corresponding spectral shape preservation coefficient can be corrected to zero, so that the spectral shape preservation coefficient can be used as a non-negative response in the subsequent shading darkening response calculation. The spectral shape preservation coefficient is used because ridge shading usually manifests as an overall reduction in multiple bands, but the shape of the spectral curve is relatively preserved. In contrast, the real wet area not only shows a reduction in reflectivity, but also produces absorption shape changes in the moisture-related bands. Therefore, by using both the degree of darkening and the degree of spectral shape preservation, the shading darkening area can be located more accurately. The darkening level is fused with the spectral shape preservation coefficient to obtain the shading darkening response:

[0033] in, Indicates pixel position The shading and darkening response at the location; Indicates the degree of darkness of the pixel; This represents the spectral shape preservation coefficient of the pixel. Product fusion is used because the shaded darkening area of ​​the hillside should simultaneously meet the two conditions of low brightness and relatively preserved spectral shape. If a pixel only has low brightness but its spectral shape has a significant absorption change, it is more likely to belong to a real wet area or other material changes, and should not be identified as the shaded darkening center alone.

[0034] It should be noted that the shading darkening response in this embodiment refers to an image response quantity used to characterize the degree to which a pixel is affected by the geometric shading of the ridge slope. This response quantity is jointly determined by the darkening degree within the same cross-section and the degree of preservation of multi-band spectral shape. The darkening degree reflects the reduction in brightness of the pixel relative to other pixels in the current cross-section, while the degree of preservation of multi-band spectral shape reflects whether the pixel exhibits an overall reduction in multiple bands while maintaining a relatively stable spectral shape. A larger shading darkening response indicates that the corresponding pixel is more likely to belong to a shading darkening area formed by insufficient light received by the ridge slope, rather than a localized area of ​​enhanced absorption caused by actual wetting absorption.

[0035] Extract the location of the maximum response to shading and darkening on each cross section:

[0036] in, Indicates the position along the ridge direction The shading and darkening center position on the cross-section; Indicates the pixel position in this cross-section The shading and darkening response at the location; Represents the coordinates in the cross-sectional direction The search algorithm finds the coordinates that maximize the shadow darkening response. If multiple similar maximum response locations exist on the same cross-section, the weighted center of the response can be taken as the shadow darkening center to reduce center jumps caused by noisy pixels. The shadow darkening center positions on each cross-section are then aligned with... Connect the directions to obtain the center line for shading and darkening.

[0037] It should be noted that the shading darkening center line in this embodiment refers to the sequence of center positions formed by sequentially connecting the positions where the shading darkening response reaches its maximum or is most representative on each cross-section along the ridge direction. This center line is used to represent the spatial extension of the shading darkening area of ​​the ridge slope along the ridge direction and serves as a benchmark for subsequent cross-sectional position comparison with the wetting absorption center line. By extracting the shading darkening center line, the continuous darkening zone formed by the backlit ridge slope can be separated from the ordinary low-brightness area. The shading darkening response obtained in this step is subsequently used to construct the shading influence distribution, and the shading darkening center line is subsequently used to calculate the center misalignment with the wetting absorption center line and participate in the shading suppression correction of the soil moisture state index.

[0038] In one embodiment of the present invention, envelope absorption analysis is performed on the cross-sectional coordinate image of the ridge, and the wetting absorption response is determined based on the difference in the obtained envelope absorption response in the image space. The wetting absorption centerline is then extracted from the cross-section, including: Specifically, continuous envelope processing is performed on the spectral curve of each pixel in the cross-sectional coordinate image of the ridge to obtain the envelope reflection response:

[0039] in, Indicates pixel position The spectral curve at the band Envelope reflection response; This indicates a continuous envelope fitting operation performed on the spectral curve along the spectral band direction. This indicates the position of the pixel in the cross-sectional coordinate image of the ridge in the band. The reflection response is obtained from the local upper envelope points of the spectral curve; the continuous envelope can be formed by linear interpolation or piecewise fitting. Its function is to construct the background reflection trend of the spectral curve of the same pixel, so that the local absorption valley can be quantified in the form of relative to the background trend. Specifically, when performing continuous envelope processing on the spectral curve of a single pixel, the reflection response of the pixel within the effective band is first arranged in band order, and invalid bands, missing bands, and abnormal peak bands are removed. Then, band points located on the convex boundary of the spectral curve in the remaining bands are selected as envelope support points. Linear interpolation or piecewise fitting is used between adjacent envelope support points to form the envelope reflection response. When the envelope reflection response obtained by interpolation or fitting is lower than the actual reflection response of the corresponding band, the envelope reflection response of that band is corrected to be no lower than the actual reflection response to avoid negative absorption depth due to insufficient envelope fitting. For regions with large band intervals between envelope support points, piecewise linear functions between adjacent envelope support points can be used for completion. For abnormal envelope support points caused by local noise peaks, they can be removed by combining the continuity of the reflection response of adjacent bands. Through the above processing, the envelope absorption response can stably characterize the local absorption depth of the spectrum, avoiding misjudgment of the wetting absorption response due to noise peaks, invalid bands, or envelope fitting errors. The envelope absorption response is obtained based on the absorption depth of the cross-sectional coordinate image of the ridge relative to the envelope reflection response:

[0040] in, Indicates pixel position In the band Envelope absorption response at the location; This indicates that the pixel is in the band. The reflection response under the following conditions; This indicates that the pixel is in the band. Envelope reflection response; This indicates a very small positive number used to avoid division by zero; the envelope absorption response is used because real soil moisture will form an enhanced absorption in the water-related band relative to the continuous envelope, while ridge shading tends to darken the overall color of multiple bands, and the two differ in their envelope absorption performance. It should be noted that the envelope absorption response in this embodiment refers to the local absorption depth formed by the pixel spectral curve relative to its continuous envelope reflectance trend. The continuous envelope reflectance response is used to represent the background reflectance trend of the pixel in each band, and the envelope absorption response is used to represent the degree of decrease in the actual reflectance response relative to the background reflectance trend. Unlike directly using reflectance reduction, the envelope absorption response can highlight local absorption changes in the shape of the spectral curve, and is therefore suitable for distinguishing between absorption enhancement caused by actual wetting and overall darkening of multiple bands caused by shading from hillsides.

[0041] Within the effective band range of soil moisture absorption changes, the band with the greatest spatial difference in envelope absorption response in the current image is identified as the moisture response band:

[0042] in, This indicates the wetting response band in the current hyperspectral image used to characterize changes in soil wetting absorption; The set of effective bands representing changes in soil moisture absorption; This represents the set of valid farmland pixels in the coordinate image of the cross-section of the ridge. Indicates band Lower envelope absorption response in image space coordinates variance on; Indicates the envelope absorption response; The effective band set The set of bands in hyperspectral remote sensing images used for moisture absorption response analysis is determined through a process that includes: first, eliminating edge bands where the sensor response is below the effective imaging requirements, abnormal bands with stripe noise or saturated response, bands with a missing pixel ratio that does not meet the imaging requirements, and bands that cannot stably reflect changes in surface reflection due to strong atmospheric water vapor absorption; then, retaining continuous bands that can reflect changes in soil moisture absorption from the remaining bands. For UAV hyperspectral images, the effective band set can be determined based on the effective response range of the acquisition equipment, whiteboard correction results, and the noise level of each band in the farmland area; for airborne or satellite hyperspectral images, the effective band set can be determined based on the effective band identifiers, atmospheric correction quality identifiers, and band stability in the farmland area from the sensor product. Moisture response bands are adaptively determined within the available image bands, rather than being mistakenly selected from invalid bands, high-noise bands, or bands significantly affected by atmospheric interference. The band with the greatest spatial difference is used as the wet response band because the actual wet area after rainfall or irrigation is locally enhanced in the furrow, the lower part of the ridge slope or near the infiltration front, which can form a large spatial difference in the corresponding band. However, using a fixed band may be affected by the sensor band setting, soil type or imaging conditions, which is not conducive to stable implementation in different scenarios.

[0043] Construct the wetting absorption response based on the wetting response band:

[0044] in, Indicates pixel position Moisture absorption response at the location; This indicates that the pixel is in the wet response band. Envelope absorption response at the location; This represents the wetting response band determined by the principle of maximizing spatial difference. A larger wetting absorption response indicates that the pixel is more likely to have enhanced absorption caused by an increase in actual soil moisture. It should be noted that the wetting absorption response in this embodiment refers to an image response quantity used to characterize changes in soil moisture absorption in a pixel. This response quantity is determined by the envelope absorption response at the wetting response band and reflects the degree of absorption enhancement related to soil moisture in the pixel's spectral curve. A larger wetting absorption response indicates that the corresponding pixel is more likely to experience absorption enhancement caused by increased soil moisture. This wetting absorption response is not equivalent to a normal decrease in brightness, but rather is used to characterize the absorption change in the spectral shape relative to the envelope reflection trend.

[0045] Extract the location of the maximum response to wetting absorption on each cross section:

[0046] in, Indicates the position along the ridge direction The location of the moisture absorption center on the cross-section; Indicates the pixel position in this cross-section Moisture absorption response at the location; Represents the coordinates in the cross-sectional direction The search is performed to find the coordinates that maximize the wetting absorption response. If the wetting absorption response exhibits a plateau-like distribution in a local region, the weighted center of the local response can be used as the wetting absorption center to improve the continuity of the centerline. The positions of the wetting absorption centers on each cross-section are then aligned with... Connect the directions to obtain the center line of moisture absorption; It should be noted that the wetting absorption centerline in this embodiment refers to the sequence of central positions formed by connecting the locations where the wetting absorption response reaches its maximum or is most representative on each cross-section along the ridge direction. This centerline is used to represent the spatial extension of the wetting absorption area along the ridge direction and to compare its position in the cross-sectional direction with the shading darkening centerline. By extracting the wetting absorption centerline, the subsequent correction process no longer simply relies on low-reflection dark areas, but instead corrects the soil moisture state based on the locations with stronger actual wetting absorption responses. The wetting absorption response output in this step is used to subsequently construct the wetting influence distribution and standardize the wetting absorption response, while the wetting absorption centerline is used to subsequently calculate the misalignment stability weight and participate in the wetting reinforcement calculation.

[0047] In one embodiment of the present invention, the misalignment stability weight is calculated based on the relative misalignment relationship and the degree of misalignment fluctuation between the shading and darkening center line and the moisture absorption center line in the cross-sectional direction, including: Specifically, when extracting the shading darkening center line and the wetting absorption center line, the validity of each cross section is determined. If the number of effective pixels in the cross section is insufficient, or if there is no distinguishable local enhancement region in the shading darkening response within the cross section, then the cross section is not used to extract the shading darkening center position. Similarly, if there is no distinguishable local enhancement region in the wetting absorption response within the cross section, then the cross section is not used to extract the wetting absorption center position. Cross sections where the shading darkening center position or the wetting absorption center position cannot be reliably extracted are marked as invalid cross sections, and invalid cross sections do not participate in the calculation of center misalignment and misalignment stability weights. For short-distance invalid cross sections between adjacent effective cross sections, linear interpolation can be performed to complete the cross section based on the center positions of adjacent effective cross sections to maintain the continuity of the center line. For regions with long continuous invalid cross sections, no correction based on the center misalignment relationship is performed; only the standardized wetting absorption response is retained for subsequent state division. Through the above processing, it is possible to avoid center line jumps caused by vegetation residue, missing pixels, local strong reflections, or local noise, and to prevent abnormal center lines from participating in the calculation of misalignment stability weights.

[0048] Furthermore, based on the relative misalignment relationship and degree of misalignment fluctuation between the shading and darkening center lines and the moisture absorption center lines in the cross-sectional direction, the misalignment stability weight is calculated; for the same position along the ridge direction... Calculate the center misalignment of the moisture absorption centerline relative to the shading and darkening centerline at the cross-section:

[0049] in, Indicates the position along the ridge direction The center misalignment at the location; This indicates the location of the wetting absorption center on the cross-section; This indicates the location of the shading and darkening center on the cross-section. The difference between the two is used to calculate the center misalignment because this embodiment focuses on the offset direction and offset distance of the actual wetting absorption center relative to the shading and darkening center in the cross-sectional direction of the ridge. This misalignment can directly characterize whether there is a stable spatial deviation between the dark zone of the backlit ridge slope and the actual wetting area. It should be noted that the center misalignment amount in this embodiment refers to the offset of the position of the wetting absorption center line relative to the shading and darkening center line in the cross-sectional direction within the same ridge cross-section. This offset amount is used to describe the relative positional relationship between the center of the shading and darkening area and the center of the wetting absorption area. The center misalignment amount can reflect both whether the two coincide and the offset direction and distance of the wetting absorption center relative to the shading and darkening center. It is the basic data for subsequent judgment on whether the misalignment relationship is stable and whether correction processing is required. Before calculating the median value of center misalignment, the set of effective cross sections for misalignment statistics is first determined. This set of effective cross sections consists of cross sections that simultaneously possess both a reliable shading and darkening center position and a reliable wetting absorption center position; cross sections marked as invalid are not included in the center misalignment statistics. If the number of effective cross sections is insufficient to reflect stable misalignment relationships along the ridge direction, the misalignment stability weight is reduced or no correction processing based on center misalignment relationships is performed, ensuring that the subsequent soil wetting state index is primarily determined by the standardized wetting absorption response. By limiting the statistical objects of center misalignment, interference from invalid cross sections, noisy cross sections, or centerline jump cross sections can be avoided on the misalignment stability weight.

[0050] Calculate the median value of the center misalignment corresponding to different effective cross sections:

[0051] in, This represents the median misalignment value obtained statistically along the ridge direction; Indicates the position along the ridge direction The center misalignment at the location; This represents the set of valid cross-sections participating in misalignment statistics. This indicates that the median value of the center misalignment is taken along the effective cross-section set. The median value is used to reduce the impact of the extraction deviation of the shading center or the moist center in a few abnormal cross sections on the judgment of the overall misalignment direction and misalignment distance. Calculate the absolute deviation of each center misalignment relative to the median misalignment value, and use the median of this absolute deviation as the degree of misalignment fluctuation:

[0052] in, Indicates the degree of misalignment fluctuation; Indicates the position along the ridge direction The center misalignment at the location; This represents the midpoint misalignment value of the center misalignment. This represents the set of effective cross sections participating in the misalignment statistics. The median value of the absolute deviation is used because a stable misalignment should be characterized by a nearly uniform amount of misalignment along the ridge direction. If the amount of misalignment fluctuates greatly on different cross sections, it indicates that there is no stable correspondence between the shading and darkening center and the moisture absorption center. In this case, it is not advisable to carry out strong correction. Calculate the misalignment stability weight based on the absolute value of the median of the central misalignment and the degree of misalignment fluctuation:

[0053] in, Indicates misaligned stable weights; This represents the midpoint misalignment value of the center misalignment. Indicates the degree of misalignment fluctuation; This represents a very small positive number used to avoid division by zero; the above calculation method is used because when the shading and darkening center and the moisture absorption center have a significant and stable misalignment in the cross-sectional direction... Larger and Smaller, calculated A larger value indicates that shading suppression and moisture reinforcement can be effectively implemented; when the misalignment between the two is not obvious or the fluctuation along the ridge direction is large, A smaller value indicates that corrections should not be made excessively based on the central misalignment relationship; this misalignment stabilization weight is subsequently used to control the correction intensity of the soil moisture state index, avoiding erroneous corrections in areas where the misalignment relationship is unstable.

[0054] It should be noted that the misalignment stability weight in this embodiment refers to the weight calculated based on the central misalignment amount and its fluctuation degree on different cross-sections. It is used to characterize whether the misalignment relationship between the shading darkening center line and the moisture absorption center line is stable along the ridge direction. When the central misalignment amount has a relatively consistent offset direction and offset distance along the ridge direction, the misalignment stability weight is larger, indicating that shading suppression and moisture reinforcement can be strongly based on the central misalignment relationship. When the central misalignment amount is not significant or fluctuates greatly along the ridge direction, the misalignment stability weight is smaller, indicating that excessive correction based on the central misalignment relationship is not advisable. This weight is used to control the correction intensity of the subsequent soil moisture state index.

[0055] In one embodiment of the present invention, a modified soil moisture state index is generated based on misalignment stabilization weights, shading darkening response, and wetting absorption response, including: Calculate the cross-sectional distribution width around the shading centerline based on the shading darkening response:

[0056] in, Indicates the position along the ridge direction The width of the cross-sectional distribution of the shading and darkening response around the center line of the shading and darkening; Indicates the cross-sectional direction coordinates; Indicates the position along the ridge direction The set of effective pixels within the cross-section; This indicates the position of the shading / darkening center line on this cross section; Indicates pixel position The shading and darkening response at the location; This represents a very small positive number used to avoid division by zero; the second-order distribution width, which is weighted by the response, is used to adaptively determine the shading influence range based on the expansion range of the shading darkening response itself in the cross section. Based on this cross-sectional distribution width, a shading effect distribution reflecting the degree to which pixels approach the shading darkening center line is constructed:

[0057] in, Indicates pixel position The distribution value of the shading effect at the location; Indicates the cross-sectional direction coordinates; Indicates the position of the shading and darkening center line along the ridge direction. The cross-sectional position at that location; This indicates the width of the shading and darkening response distribution on this cross section; Represents an exponential function; This represents a very small positive number used to avoid division by zero. The reason for using this form to construct the shading effect distribution is that the closer a pixel is to the shading darkening center, the higher the probability that it will be affected by the ridge shading darkening effect. As the cross-sectional distance from the shading darkening center increases, the shading effect should gradually weaken. It should be noted that the shading effect distribution in this embodiment refers to a spatial distribution constructed based on the cross-sectional distance between the pixel and the shading darkening center line and the cross-sectional distribution width of the shading darkening response. This distribution represents the degree to which a pixel is close to the shading darkening center line. The closer a pixel is to the shading darkening center line, the larger its shading effect distribution value, indicating that the pixel is more likely to be affected by the ridge shading darkening; the farther a pixel is from the shading darkening center line, the smaller its shading effect distribution value, indicating that the shading darkening effect gradually weakens. This shading effect distribution is used to subsequently determine which pixels' wetting responses need to be suppressed.

[0058] Similarly, the cross-sectional distribution width around the moisture absorption centerline is calculated based on the moisture absorption response:

[0059] in, Indicates the position along the ridge direction The width of the cross-sectional distribution of the wetting absorption response around the center line of the wetting absorption; Indicates the cross-sectional direction coordinates; This represents the set of valid pixels within the cross-section; This indicates the position of the wetting absorption centerline on this cross-section; Indicates pixel position Moisture absorption response at the location; This represents a very small positive number used to avoid division by zero; the distribution width weighted by the wetting absorption response is used to make the range of wetting influence adaptively change with the spatial extent of the actual wetting absorption response. Based on this cross-sectional distribution width, a wetting effect distribution reflecting the degree to which a pixel approaches the center line of wetting absorption is constructed:

[0060] in, Indicates pixel position The humidity at a given location affects the distribution values; Indicates the cross-sectional direction coordinates; Indicates the position of the moisture absorption center line along the ridge direction. The cross-sectional position at that location; This indicates the width of the wetting absorption response distribution across the cross-section; Represents an exponential function; This represents a very small positive number used to avoid division by zero; the wetting effect distribution is used to preserve and enhance pixels close to the true wetting absorption center in subsequent corrections, rather than simply suppressing all low-brightness areas. It should be noted that the wetting effect distribution in this embodiment refers to a spatial distribution constructed based on the cross-sectional distance between the pixel and the wetting absorption center line and the cross-sectional distribution width of the wetting absorption response. This distribution represents the degree to which a pixel is close to the wetting absorption center line. The closer a pixel is to the wetting absorption center line, the larger its wetting effect distribution value, indicating that the pixel is more likely to be near the actual wetting absorption area; the farther a pixel is from the wetting absorption center line, the smaller its wetting effect distribution value, indicating that the influence of the wetting absorption center on the pixel gradually weakens. This wetting effect distribution is used to subsequently retain and reinforce the wetting response close to the wetting absorption center line.

[0061] The moisture absorption response was robustly normalized to obtain the standardized moisture absorption response:

[0062] in, Indicates pixel position Standardized wetting absorption response at the location; Indicates pixel position Moisture absorption response at the location; This represents the median value of the moisture absorption response within the entire cross-sectional coordinate image of the ridge or the effective farmland area; This represents the median absolute deviation of the wet absorption response within the entire cross-sectional coordinate image of the ridge or the effective farmland area. This indicates a very small positive number used to avoid division by zero; robust normalization is employed to convert the wetting absorption response of different images, different soil backgrounds and different imaging conditions to a comparable scale, while reducing the impact of abnormally high absorption pixels or noisy pixels on the overall wetting status judgment.

[0063] After obtaining the misalignment stabilization weights, shading effect distribution, wetting effect distribution, and standardized wetting absorption response, the inhibition factor is first calculated:

[0064] in, Indicates pixel position Inhibitory factors at the location; Indicates misaligned stable weights; This indicates the distribution value of the shading effect; This represents the distribution value of the wetting effect. This suppression factor is used because a pixel's wetting response should only be considered more likely to be affected by shading and suppressed if it is close to the shading darkening center and far from the wetting absorption center. If a pixel is close to the wetting absorption center, its wetting absorption contribution should be preserved even if it is in a low-brightness area.

[0065] Calculate the base response retention value:

[0066] in, Indicates pixel position The base response value is retained at the location; This indicates the suppression factor for that pixel; This represents the standardized wetting absorption response of the pixel. The use of the base response retention value is to preserve the pixel's own wetting absorption information while suppressing shading-induced misleading responses, thus avoiding the creation of recognition holes by directly deleting local areas. Calculate the center wetting reinforcement value:

[0067] Where, represents the pixel position. The center wetting reinforcement value at the location; Indicates misaligned stable weights; This indicates the distribution value affected by humidity; This represents the standardized wetting absorption response at the current wetting absorption centerline position on the cross-section; if The coordinates are not integer pixels and can be obtained through linear interpolation of adjacent pixels. The center wetting reinforcement value is used because the true wetting absorption center represents the most reliable wetting absorption position on the cross-section. When the misalignment is stable, the wetting state identification result should be regressed to the vicinity of the true wetting absorption center.

[0068] The soil moisture state index after correction for the current pixel is obtained by adding the base response retention value to the center moisture reinforcement value:

[0069] in, Indicates pixel position The corrected soil moisture index; Indicates the base response reserved value; This indicates the center wetting reinforcement value; the above steps can also be combined and expressed as:

[0070] in, This indicates the corrected soil moisture index; Indicates pixel position Standardized wetting absorption response at the location; Indicates misaligned stable weights; This indicates the distribution value of the shading effect; This indicates the distribution value affected by humidity; This represents the standardized wetting absorption response at the current wetting absorption centerline position on the cross-section; This indicates the location of the wetting absorption center on the current cross-section. The technical implication of this calculation method is that when a pixel is close to the shading darkening center and far from the wetting absorption center... As the pixel size increases, its wetting absorption response is suppressed; when the pixel is close to the wetting absorption center, The increase in the central wetting absorption response is used to reinforce the area, thereby causing the soil wetting state index to revert from the shading and darkening zone of the ridge slope to the true wetting absorption area. In this way, this embodiment does not simply identify the dark area as a wet area, but corrects the position of the wetting response based on the stable misalignment relationship between the shading and darkening center and the wetting absorption center. It should be noted that the corrected soil moisture state index in this embodiment refers to the image state quantity obtained by combining misalignment stability weights, shading influence distribution, and moisture influence distribution with the standardized moisture absorption response to suppress shading and enhance moisture. This state index is used to characterize the dry and wet state of farmland soil after correction for the effects of shading and darkening on the ridge slope. When a pixel is close to the shading and darkening center line and far from the moisture absorption center line, the moisture response of that pixel will be suppressed; when a pixel is close to the moisture absorption center line, the moisture response of that pixel or its neighboring area will be preserved and enhanced. The corrected soil moisture state index can reduce the misleading effect of backlit ridge slope darkening on soil moisture identification, making the final soil state identification result closer to the true moisture absorption area.

[0071] In one embodiment of the present invention, ordered state division is performed based on the modified soil moisture state index to obtain the classification result under the ridge coordinate system, and the classification result is inversely transformed to the original image coordinate system to obtain a farmland soil state identification map, including: Specifically, first determine the number of soil condition levels. Number of soil condition grades It can be determined based on agricultural management needs, irrigation control granularity, or sample calibration results. For example, It can be divided into two, three or more categories. Category two can represent dry and wet areas, category three can represent dry, transitional and wet areas, and multiple categories can be used to represent more detailed soil moisture levels. Based on the number of soil condition levels, the corrected soil moisture state index is clustered and classified into ordered states:

[0072] in, Indicates the number of soil condition levels; This represents the j-th soil condition category; j represents the index of the soil condition category. Indicates pixel position The corrected soil moisture index; This represents the mean of the wet state corresponding to the j-th soil state category; Indicates pixel position Soil is classified into the j-th soil condition category. The purpose of this clustering objective is to make the soil moisture state indices within the same category as close as possible, so that different categories can reflect different moisture levels. Specifically, since the corrected soil moisture state index is a single continuous state quantity, one-dimensional clustering can be used to cluster and classify its ordered states. First, the mean moisture state value for each soil state category is initialized based on the numerical distribution of the corrected soil moisture state index. Then, pixel classification and category mean update operations are repeated until the change in the mean moisture state value for each category meets the convergence requirement. Pixel classification refers to assigning each pixel to the soil state category closest to its soil moisture state index, and category mean update refers to recalculating the mean moisture state value for that category based on pixels assigned to the same category. Using one-dimensional clustering can directly classify states from drier to wetter based on the soil moisture state index, avoiding the introduction of additional classification features unrelated to the misalignment between shading and moisture absorption centers.

[0073] Soil condition categories were sorted in ascending order of mean moisture content, resulting in a classification of soil conditions from drier to wetter in the ridge coordinate system. ,

[0074] in, This represents the soil state classification results in the coordinate system of the ridge cross section; This represents the average value according to each category of humidity. The category hierarchy arranged in ascending order; This represents the j-th soil state category. The ordered arrangement is used because the magnitude of the wetness index corresponds to the trend of soil change from dry to wet. Directly sorting the categories can give the output results a clear agricultural interpretation. The classification results in the ridge coordinate system are subjected to an inverse transformation from the ridge coordinate system to the original image coordinate system:

[0075] in, A farmland soil condition recognition map representing the original image coordinate system; This represents the horizontal pixel coordinates in the original image coordinate system; Represents the vertical pixel coordinates in the original image coordinate system; Representation and coordinate transformation operations The corresponding inverse transform operation; This represents the soil state classification results in the coordinate system of the ridge cross section; It indicates the direction of the identified ridges; through inverse transformation, the identification results under the ridge coordinates can be mapped back to the original hyperspectral remote sensing image, so that the output farmland soil condition identification map is consistent with the spatial position of the original image, which is convenient for subsequent irrigation management, farmland monitoring layer overlay or operation area positioning. It should be noted that the farmland soil condition identification map in this embodiment refers to the image result formed after inversely transforming the soil condition classification results obtained in the ridge cross-section coordinate system to the original image coordinate system. The pixel positions in this identification map correspond to the spatial positions of the original hyperspectral remote sensing image, and the pixel values ​​are used to represent the soil condition category at that location, such as a slightly dry area, a transitional area, or a slightly wet area. By outputting the farmland soil condition identification map, the soil condition results obtained in this embodiment can be used for farmland irrigation management, crop condition monitoring layer overlay, and operation area positioning.

[0076] In one optional embodiment, the hyperspectral remote sensing image can be acquired by a UAV hyperspectral camera, or by an airborne or satellite hyperspectral sensor. The acquired image is processed by radiometric correction, geometric correction, farmland boundary cropping, and invalid pixel masking before being input into the method of this embodiment. In one optional embodiment, the ridge direction can be determined by the directional correlation of the ridge texture brightness map, or by the direction of the strip texture formed by the furrows and ridge tops in the image. Any image that can be converted into a ridge cross-sectional coordinate image with the cross-sectional direction and the ridge direction as coordinates can be used in this embodiment. In one optional embodiment, the wetting response band is not limited to a fixed band, but is adaptively determined by the spatial differences in the envelope absorption response to adapt to different sensor band settings, different soil types, and different imaging conditions, avoiding unstable recognition results due to the failure of a fixed band. In one optional embodiment, the number of soil state levels can be set to two, three, or more categories according to application requirements. Different numbers of levels will not change the core processing logic of shading darkening center extraction, wetting absorption center extraction, misalignment stabilization weight calculation, and soil wetting state index correction.

[0077] It should be noted that the adjacent displacement, effective band set, shading influence distribution width, wetting influence distribution width, and number of soil state levels in this embodiment can be adaptively determined based on the spatial texture, band quality, response distribution, and agricultural management needs of the input hyperspectral remote sensing image, or determined according to application requirements, and are not limited to fixed values. The determination of the above parameters serves the identification and correction of the misalignment between the shading darkening center and the wetting absorption center in the cross-sectional direction of the ridge, and will not change the core technical logic of this embodiment of correcting the soil wetting state index through the center misalignment relationship.

[0078] like Figure 2 The diagram shown is a functional block diagram of a farmland soil condition identification system based on hyperspectral remote sensing image processing provided in an embodiment of the present invention.

[0079] In this embodiment, the functions of each module / unit are as follows: The image coordinate transformation unit is used to acquire hyperspectral remote sensing images of farmland areas, determine the ridge direction based on the ridge texture in the hyperspectral remote sensing images, and transform the hyperspectral remote sensing images to a coordinate system with the ridge direction perpendicular to the ridge direction as the cross-sectional direction, to obtain a ridge cross-sectional coordinate image. The shading center extraction unit is used to determine the shading darkening response based on the degree of darkening in the cross-sectional coordinate image of the ridge and the similarity of the multi-band spectral shape, and to extract the shading darkening center line on the cross-section. The wetting center extraction unit is used to perform envelope absorption analysis on the cross-sectional coordinate image of the ridge, and determine the wetting absorption response based on the difference of the obtained envelope absorption response in the image space, and extract the wetting absorption center line on the cross-section. The misalignment weight calculation unit is used to calculate the misalignment stability weight based on the relative misalignment relationship and the degree of misalignment fluctuation between the shading and darkening center line and the moisture absorption center line in the cross-sectional direction. The index correction unit is used to generate a corrected soil moisture state index based on the misalignment stability weight, shading darkening response, and wetting absorption response. The state recognition unit is used to divide the soil into ordered states according to the corrected soil moisture state index, obtain the classification result in the ridge coordinate system, and then transform the classification result back to the original image coordinate system to obtain the farmland soil state recognition map.

[0080] This embodiment also provides a computer device, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor. The memory can be used to store hyperspectral remote sensing images, farmland area masks, normalization parameters, ridge direction parameters, ridge cross-sectional coordinate images, shading darkening response, moisture absorption response, shading darkening centerline, moisture absorption centerline, misalignment stabilization weights, corrected soil moisture state index, and farmland soil state identification maps. When the processor executes the computer program, it sequentially performs the following steps: acquisition of hyperspectral remote sensing images of farmland areas with ridge texture, robust normalization processing, ridge direction determination, ridge cross-sectional coordinate transformation, shading darkening response calculation, shading darkening centerline extraction, envelope absorption analysis, moisture absorption response calculation, moisture absorption centerline extraction, center misalignment calculation, misalignment stabilization weight calculation, shading influence distribution construction, moisture influence distribution construction, standardized moisture absorption response calculation, soil moisture state index correction, ordered state partitioning, and coordinate inverse transformation output. The computer equipment can be a computing device installed in a UAV remote sensing data processing platform, an agricultural remote sensing monitoring server, an edge computing terminal, or a general image processing workstation.

[0081] like Figure 3As shown, ridge-cultivated farmland consists of multiple ridges extending in the same direction. Each ridge includes a ridge top 1, a ridge slope 2, and a furrow 3. The direction of solar incidence causes differences in light exposure on different slopes of the ridge. Shaded and darkened areas 4 are formed on the shaded ridge slopes, while truly moist areas 5 are formed in the furrows 3 or the lower part of the ridge slope 2 after rainfall or irrigation. A magnified section 6 shows that the center of the shaded and darkened area 4 does not coincide with the center of the truly moist area 5 in the cross-sectional direction of the ridge; there is an offset 7 between them. This causes low-reflectance darkened areas in hyperspectral remote sensing images to be easily mistaken for moist soil areas, resulting in a misidentification of the soil moisture status in farmland. This invention addresses this scenario by extracting the shaded and darkened center lines and the moist absorption center lines separately, and correcting the soil moisture status index based on the cross-sectional misalignment relationship between the two, thereby reducing the interference of ridge slope shading and darkening on the farmland soil status identification results.

[0082] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for identifying farmland soil conditions based on hyperspectral remote sensing image processing, characterized in that, include: S1. Obtain a hyperspectral remote sensing image of the farmland area, determine the ridge direction based on the ridge texture in the hyperspectral remote sensing image, and convert the hyperspectral remote sensing image to a coordinate system with the ridge direction perpendicular to the ridge direction as the cross-sectional direction to obtain a ridge cross-sectional coordinate image. S2. Based on the degree of darkening in the cross-sectional coordinate image of the ridge and the similarity of the multi-band spectral shape, determine the shading darkening response and extract the shading darkening center line on the cross-section. S3. Perform envelope absorption analysis on the cross-sectional coordinate image of the ridge, and determine the wetting absorption response based on the difference in the obtained envelope absorption response in the image space, and extract the wetting absorption center line on the cross-section. S4. Calculate the misalignment stability weight based on the relative misalignment relationship and degree of misalignment fluctuation between the shading and darkening center line and the moisture absorption center line in the cross-sectional direction. S5. Based on the misaligned stability weights, shading darkening response, and wetting absorption response, a corrected soil wetting state index is generated. S6. Based on the corrected soil moisture state index, the soil is divided into ordered states to obtain the classification results under the ridge coordinates. The classification results are then transformed back to the original image coordinate system to obtain the farmland soil state recognition map.

2. The method for identifying farmland soil conditions based on hyperspectral remote sensing image processing according to claim 1, characterized in that, Determining the ridge direction based on the ridge texture in the hyperspectral remote sensing image, and transforming the hyperspectral remote sensing image to a coordinate system with a cross-sectional direction perpendicular to the ridge direction, includes: Robust normalization processing is performed on the reflection response of each band in the acquired hyperspectral remote sensing image; Based on the normalized reflection response of each band, the multi-band comprehensive brightness response is calculated, and the ridge texture brightness map is constructed. The brightness map of the ridge texture is sampled or projected along different candidate directions to form a brightness profile of the candidate direction. Calculate the ridge repeat correlation intensity of the candidate direction brightness profile under adjacent displacements, and determine the candidate direction that produces the maximum ridge repeat correlation intensity as the ridge direction; Using the ridge direction as the longitudinal direction and the direction perpendicular to the ridge direction as the cross-sectional direction, coordinate transformation is performed on the normalized hyperspectral remote sensing image to obtain the ridge cross-sectional coordinate image.

3. The method for identifying farmland soil conditions based on hyperspectral remote sensing image processing according to claim 1, characterized in that, Based on the degree of darkening in the cross-sectional coordinate image of the ridge and the similarity of its multi-band spectral shape, the shading darkening response is determined, and the shading darkening center line is extracted on the cross-section, including: Calculate the multi-band composite brightness of pixels in the cross-sectional coordinate image of the ridge; The degree of darkening is calculated based on the relative relationship between the multi-band combined brightness of pixels within the same cross-section and the minimum and maximum multi-band combined brightness within that cross-section. The correlation between the normalized spectral shape of the current pixel and the average spectral shape of the current cross-section is calculated to obtain the spectral shape preservation coefficient; The darkening level is fused with the spectral shape preservation coefficient to obtain the shading darkening response; Extract the location of the maximum response of the shading darkening on each cross section, and connect the locations of the maximum response on each cross section to form the shading darkening center line.

4. The method for identifying farmland soil conditions based on hyperspectral remote sensing image processing according to claim 1, characterized in that, Envelope absorption analysis was performed on the cross-sectional coordinate image of the ridge, and the wetting absorption response was determined based on the differences in the obtained envelope absorption response in the image space. The wetting absorption centerline was then extracted from the cross-section, including: The spectral curves of each pixel in the cross-sectional coordinate image of the ridge are subjected to continuous envelope processing to obtain the envelope reflection response; The envelope absorption response is obtained based on the absorption depth of the cross-sectional coordinate image of the ridge relative to the envelope reflection response. Within the effective band range of soil moisture absorption changes, the band with the greatest spatial difference in envelope absorption response in the current image is identified as the moisture response band. The envelope absorption response corresponding to the wetting response band is taken as the wetting absorption response; The location of the maximum response to the wetting absorption is extracted on each cross-section, and the locations of the maximum response on each cross-section are connected to form the center line of the wetting absorption.

5. The method for identifying farmland soil conditions based on hyperspectral remote sensing image processing according to claim 1, characterized in that, Based on the relative misalignment between the shading and darkening center lines and the moisture absorption center lines in the cross-sectional direction, and the degree of misalignment fluctuation, the misalignment stability weight is calculated, including: Calculate the center misalignment of the moisture absorption centerline relative to the shading and darkening centerline on the same cross-section; Calculate the median value of the center misalignment corresponding to different cross sections, calculate the absolute deviation of each center misalignment relative to the median value of the center misalignment, and use the median value of the absolute deviation as the degree of misalignment fluctuation. The misalignment stability weight is calculated based on the absolute value of the median of the central misalignment and the degree of misalignment fluctuation. Wherein, the larger the absolute value of the median value of the central misalignment and the smaller the degree of misalignment fluctuation, the larger the misalignment stability weight; the smaller the absolute value of the median value of the central misalignment or the larger the degree of misalignment fluctuation, the smaller the misalignment stability weight.

6. The method for identifying farmland soil conditions based on hyperspectral remote sensing image processing according to claim 1, characterized in that, Based on the misalignment stabilization weights, shading darkening response, and wetting absorption response, a corrected soil wetting state index is generated, including: The cross-sectional distribution width around the shading darkening center line is calculated based on the shading darkening response, and a shading effect distribution reflecting the degree to which a pixel approaches the shading darkening center line is constructed based on this cross-sectional distribution width. The cross-sectional distribution width around the wetting absorption centerline is calculated based on the wetting absorption response, and a wetting influence distribution reflecting the degree to which a pixel is close to the wetting absorption centerline is constructed based on this cross-sectional distribution width. The moisture absorption response was robustly normalized to obtain a standardized moisture absorption response. By combining the misalignment stabilization weights, the distribution of shading effects, and the distribution of wetting effects, the standardized wetting absorption response is subjected to shading suppression and wetting reinforcement calculations. This results in the suppression of pixel responses that are close to the shading darkening center line and far from the wetting absorption center line, while the reinforcement of pixel responses that are close to the wetting absorption center line, thus generating a corrected soil wetting state index.

7. The method for identifying farmland soil conditions based on hyperspectral remote sensing image processing according to claim 6, characterized in that, The method combines misalignment stabilization weights, shading effect distribution, and wetting effect distribution to perform shading suppression and wetting reinforcement calculations on the standardized wetting absorption response, including: The difference obtained by subtracting the wetting effect distribution from the baseline value of 1, multiplied by the shading effect distribution and the misalignment stability weight, yields the inhibition factor. The result of subtracting the suppression factor from the baseline value of 1 is multiplied by the normalized wetting absorption response of the current pixel to obtain the baseline response retention value. The standardized wetting absorption response at the center line of the wetting absorption cross-section of the current pixel is multiplied by the wetting influence distribution and misalignment stability weights to obtain the center wetting reinforcement value. The soil moisture state index after correction for the current pixel is obtained by adding the base response retention value and the center moisture reinforcement value.

8. The method for identifying farmland soil conditions based on hyperspectral remote sensing image processing according to claim 1, characterized in that, Based on the corrected soil moisture state index, ordered state division was performed to obtain the classification results in the ridge coordinate system. The classification results were then inversely transformed back to the original image coordinate system, including: Determine the number of soil condition grades; Based on the number of soil condition levels, the corrected soil moisture state index is clustered and classified into ordered states to obtain each soil condition category with a corresponding average moisture state value. Soil condition categories are sorted in ascending order of mean moisture content to obtain soil condition classification results under ridge coordinates. The soil state classification results under the ridge coordinate system are subjected to an inverse transformation from the ridge coordinate system to the original image coordinate system to obtain a farmland soil state identification map.

9. A farmland soil condition identification system based on hyperspectral remote sensing image processing, applied in the farmland soil condition identification method based on hyperspectral remote sensing image processing as described in any one of claims 1-8, characterized in that, include: The image coordinate transformation unit is used to acquire hyperspectral remote sensing images of farmland areas, determine the ridge direction based on the ridge texture in the hyperspectral remote sensing images, and transform the hyperspectral remote sensing images to a coordinate system with the ridge direction perpendicular to the ridge direction as the cross-sectional direction, to obtain a ridge cross-sectional coordinate image. The shading center extraction unit is used to determine the shading darkening response based on the degree of darkening in the cross-sectional coordinate image of the ridge and the similarity of the multi-band spectral shape, and to extract the shading darkening center line on the cross-section. The wetting center extraction unit is used to perform envelope absorption analysis on the cross-sectional coordinate image of the ridge, and determine the wetting absorption response based on the difference of the obtained envelope absorption response in the image space, and extract the wetting absorption center line on the cross-section. The misalignment weight calculation unit is used to calculate the misalignment stability weight based on the relative misalignment relationship and the degree of misalignment fluctuation between the shading and darkening center line and the moisture absorption center line in the cross-sectional direction. The index correction unit is used to generate a corrected soil moisture state index based on the misalignment stability weight, shading darkening response, and wetting absorption response. The state recognition unit is used to divide the soil into ordered states according to the corrected soil moisture state index, obtain the classification result in the ridge coordinate system, and then transform the classification result back to the original image coordinate system to obtain the farmland soil state recognition map.

10. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the farmland soil state identification method based on hyperspectral remote sensing image processing as described in any one of claims 1-8.

Images