A method and system for monitoring the state of a field for conservation tillage

Abstract

The application discloses a farmland state monitoring method and system for farmland protection, and relates to the technical field of image analysis.The application generates a reflectivity alignment data set through radiation normalization and cross-cycle registration, generates a vegetation confidence graph based on a vegetation discrimination band group, extracts candidate seedling center points, fuses spatial rule probability and time sequence growth probability to obtain comprehensive cultivation confidence, and then outputs cultivation boundary polygons and out-of-bound occupation polygons, and encapsulates image slices and quality scores to form an evidence package.Compared with single-time-phase and single-band interpretation, the application can resist interference of light reflection and shadow, adapt to strip sowing rules, support missing inspection, and stably identify cultivation, abandoned land and out-of-bound occupation.

Description

Technical Field

[0001] This invention relates to the field of image analysis technology, and in particular to a method and system for monitoring the status of arable land for arable land protection. Background Technology

[0002] In farmland protection, the competent authorities need to conduct routine inspections and verifications of permanent basic farmland and land parcels subject to land occupation and compensation balance, focusing on identifying situations where land that should be cultivated has not been cultivated, suspected abandoned land, suspected illegal occupation, and whether rectification and restoration have been in place. Current practices often rely on manual surveys or limited temporal remote sensing interpretations: on the one hand, due to the influence of factors such as illumination, shadows, mulch film reflection, and soil moisture, single-temporal or single-band interpretations can easily misclassify bare land, reflective areas, and field ridges as cultivated land or vegetation; on the other hand, if multi-temporal monitoring lacks reliable cross-period registration and growth consistency constraints, it is difficult to stably output boundary polygons and evidence chains that can be used for law enforcement under conditions such as grounded flights, missing data, and early crop growth stages.

[0003] Currently, Chinese invention patent application number CN202510961562.0 discloses a method for monitoring farmland status based on farmland boundary identification, which relates to the field of image analysis technology. The technical solution includes the following steps: preprocessing periodically acquired real-time farmland images to obtain grayscale images corresponding to the optimal spectral bands; calculating the comprehensive probability that each local peak pixel in the current periodic grayscale image represents a seedling based on the grayscale image and the spatial distribution and temporal difference characteristics of seedlings; determining whether each local peak pixel represents a seedling based on the comprehensive probability; and delineating farmland boundaries based on the determination results to distinguish between cultivated and uncultivated areas. This invention utilizes the spatial distribution and temporal difference characteristics of seedlings to determine the probability of seedling presence, thereby improving the accuracy of farmland boundary identification and enabling periodic real-time monitoring and updating of farmland status.

[0004] The aforementioned technologies are insufficient to simultaneously achieve accurate cross-cycle registration, suppression of complex surface interference, and law enforcement-verifiable boundary polygon output during multi-cycle drone patrols, thus making it difficult to reliably determine whether farmland is cultivated, abandoned, or occupied beyond its boundaries. Summary of the Invention

[0005] The technical problem solved by this invention is that existing technologies are unable to simultaneously achieve accurate cross-cycle registration, suppression of complex surface interference, and law enforcement-verifiable boundary polygon output during multi-cycle UAV patrols, so as to reliably determine whether farmland is cultivated, abandoned, or occupied beyond its boundaries.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: A method for monitoring the status of arable land for arable land protection includes the following steps: Step S1: Retrieve historical hyperspectral image data, historical labeled sample data, and land parcel registration boundary data to determine the vegetation discrimination band group; Step S2: Collect periodic hyperspectral image data, periodic RGB image data and flight attitude positioning data. Perform orthorectification and cross-period registration based on the flight attitude positioning data to generate a reflectance aligned dataset containing geographic coordinate mapping data. Step S3: Extract pixel reflectance from reflectance aligned dataset based on vegetation discrimination band group and calculate vegetation confidence map, detect local maxima and obtain candidate patch data by region growth, and output the centroid of candidate patch data as candidate seedling center point data. Step S4: Calculate spatial pattern probability data based on the candidate seedling center point data; Step S5: Correlate the center points of multiple cycles to form center point trajectory data and calculate the temporal growth probability data. Then, merge the spatial regularity probability data and the temporal growth probability data to obtain comprehensive farming confidence data. Step S6: Extract the cultivated boundary polygon data based on the comprehensive cultivated confidence data, compare it with the land parcel registration boundary data, and output the status result data and evidence package data.

[0007] Preferably, step S1 includes the following sub-steps: Step S101: Retrieve historical hyperspectral image data and historical labeled sample data from the database. The historical labeled sample data includes seedling sample data and non-seedling sample data. Step S102: Based on seedling sample data and non-seedling sample data, enumerate candidate band combinations in historical hyperspectral image data and calculate the band separability score data of each candidate band combination. The band separability score data is determined by the inter-class difference and the intra-class dispersion. Step S103: Select the candidate band combination with the largest band separability score data as the planting discrimination band group and store it.

[0008] Preferably, step S2 includes the following sub-steps: Step S201: Perform radiometric normalization processing on periodic hyperspectral image data and flight attitude positioning data to output periodic reflectance image data; Step S202: Orthorectify the periodic reflectivity image data based on the flight attitude positioning data, and perform feature matching and registration between periodic reflectivity image data of adjacent periods to obtain consistent spatial coordinates across periods. Step S203: Align the registered periodic reflectance image data with the periodic RGB image data in terms of coordinates, and output the reflectance aligned dataset.

[0009] Preferably, the vegetation confidence map is generated in step S3 as follows: Pixel vegetation feature vector data is constructed based on pixel reflectance data corresponding to the vegetation discrimination band group in the reflectance alignment dataset. Based on the statistical analysis of seedling sample data, seedling pattern template data was obtained, and the pattern similarity data between pixel vegetation feature vector data and seedling pattern template data was calculated. The spectral similarity data are normalized and output as a vegetation confidence map.

[0010] Preferably, step S3 includes the following sub-steps: Step S301: Detect the location of local maxima in the vegetation confidence map and output the initial center point data; Step S302: Using the initial center point data as seeds, perform region growth in the reflectance aligned dataset to obtain candidate patch data, and calculate the patch spectral consistency data and patch shape feature data corresponding to the candidate patch data; Step S303: Based on the patch pattern consistency data and patch shape feature data, remove non-seedling candidate patch data, and output the centroid of the remaining candidate patch data as the candidate seedling center point data.

[0011] Preferably, step S4 includes the following sub-steps: Step S401: Perform directional statistics on the center point data of candidate seedlings and fit the data to obtain the main sowing direction data; Step S402: Statistically analyze the spacing distribution of adjacent candidate seedling center point data along the main sowing direction, and output the main spacing data; Step S403: Calculate spatial pattern probability data based on the deviation of the candidate seedling center point data from the sowing main direction data and the spacing deviation from the main spacing data.

[0012] Preferably, step S5 includes the following sub-steps: Step S501: The candidate seedling center point data of each period are stored in partitions according to the land parcel registration boundary data to form a time-series center point dataset; Step S502: Perform cross-cycle correlation based on the time series center point dataset, output center point trajectory data, and calculate growth consistency data based on the candidate patch data corresponding to the center point trajectory data; Step S503: When a certain period is missing, interpolation is performed based on the center point trajectory data and growth consistency data of adjacent periods, and the confidence weight is reduced to output the time series growth probability data. Step S504: The spatial pattern probability data and the temporal growth probability data are fused according to weights to output comprehensive farming confidence data.

[0013] Preferably, step S6 includes the following sub-steps: Step S601: Rasterize the comprehensive cultivation confidence data within the land parcel registration boundary data and extract the connected components to obtain the cultivation connected component data; Step S602: Extract the concave contour from the cultivated connected component data and output the cultivated boundary polygon data; Step S603: Overlay the cultivated boundary polygon data with the land registration boundary data to output the over-boundary occupation polygon data, and write the cultivated boundary polygon data and the over-boundary occupation polygon data into the evidence package data.

[0014] Preferably, the logic for generating the state result data is as follows: Cultivated coverage data is calculated based on cultivated boundary polygon data and plot registration boundary data; the proportion of land occupation is calculated based on land occupation polygon data; and the risk of land abandonment is calculated based on time-series growth probability data. When the cultivated coverage data meets the first threshold and the fallow risk data meets the second threshold, the normal cultivated status is output. When the cultivated coverage data does not meet the first threshold and the fallow risk data does not meet the second threshold, output a suspected fallow status. When the out-of-bounds occupancy rate data meets the third threshold, a suspected illegal occupancy status is output. The status result data, along with the corresponding periodic RGB image data slices, reflectance aligned dataset index, cultivated boundary polygon data, and out-of-bounds occupied polygon data, are uniformly packaged and output as evidence package data.

[0015] A farmland status monitoring system for farmland protection includes a data retrieval module, a data alignment module, a center point extraction module, a pattern calculation module, a cultivation determination module, and an evidence output module. The data retrieval module is used to retrieve historical hyperspectral image data, historical labeled sample data, and land parcel registration boundary data to determine the vegetation discrimination band group; The data alignment module is used to collect periodic hyperspectral image data, periodic RGB image data and flight attitude positioning data, perform radiometric normalization and orthophoto registration, and generate reflectance aligned datasets. The center point extraction module is used to generate a vegetation confidence map based on the reflectance aligned dataset and extract candidate seedling center point data. The pattern calculation module is used to calculate spatial pattern probability data based on the candidate seedling center point data; The cultivation determination module is used to calculate time-series growth probability data based on multi-period candidate seedling center point data, and to integrate spatial regularity probability data with time-series growth probability data to obtain comprehensive cultivation confidence data. The evidence output module is used to extract cultivated boundary polygon data based on comprehensive cultivated confidence data, compare it with the land parcel registration boundary data, and output status result data and evidence package data.

[0016] The beneficial effects of this invention are as follows: This invention generates a reflectance aligned dataset through radiometric normalization and cross-period registration, generates a vegetation confidence map based on the vegetation discrimination band group and extracts the center point of candidate seedlings, integrates spatial regularity probability and temporal growth probability to obtain a comprehensive cultivation confidence score, and then outputs the cultivation boundary polygon and the over-boundary occupation polygon, and encapsulates image slices and quality scores to form an evidence package. Compared with single-temporal and single-band interpretation, it can resist reflective shadow interference, adapt to strip sowing patterns, support missing measurement inspections, and stably identify cultivation, abandonment and over-boundary occupation. Attached Figure Description

[0017] Figure 1 A flowchart illustrating the steps of a method for monitoring the status of arable land for arable land protection, as provided in one embodiment of the present invention; Figure 2 This is a basic flowchart of a farmland status monitoring system for farmland protection, provided as an embodiment of the present invention. Detailed Implementation

[0018] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0019] Example 1, referring to Figure 1 A method for monitoring the status of arable land for arable land protection is provided, comprising the following steps: Step S1: Retrieve historical hyperspectral image data, historical labeled sample data, and land parcel registration boundary data to determine the vegetation discrimination band group.

[0020] Step S2: Collect periodic hyperspectral image data, periodic RGB image data, and flight attitude positioning data. Perform orthorectification and cross-period registration based on the flight attitude positioning data to generate a reflectance aligned dataset containing geographic coordinate mapping data.

[0021] Step S3: Based on the vegetation discrimination band group, extract pixel reflectance from the reflectance aligned dataset and calculate the vegetation confidence map. Detect local maxima and perform region growth to obtain candidate patch data. Output the centroid of the candidate patch data as candidate seedling center point data.

[0022] Step S4: Calculate spatial pattern probability data based on the candidate seedling center point data.

[0023] Step S5: Correlate the center points of multiple cycles to form center point trajectory data and calculate the temporal growth probability data. Then, merge the spatial regularity probability data and the temporal growth probability data to obtain comprehensive farming confidence data.

[0024] Step S6: Extract the cultivated boundary polygon data based on the comprehensive cultivated confidence data, compare it with the land parcel registration boundary data, and output the status result data and evidence package data.

[0025] This invention generates a reflectance aligned dataset through radiometric normalization and cross-period registration, generates a vegetation confidence map based on the vegetation discrimination band group, extracts the center point of candidate seedlings, and integrates spatial regularity probability and temporal growth probability to obtain a comprehensive cultivation confidence score. It then outputs cultivation boundary polygons and over-boundary occupation polygons, and encapsulates image slices and quality scores to form an evidence package. Compared with single-temporal and single-band interpretation, it can resist reflective shadow interference, adapt to strip sowing patterns, support missing measurement inspections, and stably identify cultivation, abandonment, and over-boundary occupation.

[0026] This embodiment targets a county-level farmland protection patrol scenario, conducting multi-period monitoring of permanent basic farmland and land parcels subject to land occupation and compensation balance. It outputs cultivated boundary polygon data, boundary violation polygon data, status result data, and evidence package data. The database pre-stores parcel registration boundary data (vector polygons) and historical samples. Each cycle, a drone collects periodic hyperspectral image data, periodic RGB image data, and flight attitude positioning data, forming a consistent reflectance aligned dataset across cycles. Based on this, a vegetation confidence map and candidate seedling center point data are generated. Then, spatial pattern probability data and temporal growth probability data are calculated and fused to obtain comprehensive cultivated confidence data. Finally, cultivated boundary polygon data are extracted, and status determination is completed.

[0027] Step S1 includes the following sub-steps: Step S101: Retrieve historical hyperspectral image data and historical labeled sample data from the database. The historical labeled sample data includes seedling sample data and non-seedling sample data.

[0028] Historical hyperspectral imagery and historical labeled sample data were retrieved from the database. The historical hyperspectral imagery consisted of hyperspectral cube data cropped by plot, with each pixel containing reflectance values ​​for multiple bands. The historical labeled sample data included seedling sample data and non-seedling sample data. The seedling sample data was a set of seedling pixels manually delineated in the historical hyperspectral imagery, while the non-seedling sample data was a set of pixels manually delineated for soil, stubble, plastic film, field ridges, roads, bare soil mounds, and shadows. At the same time, plot registration boundary data was retrieved to limit the spatial range of subsequent feature statistics.

[0029] Step S102: Based on seedling sample data and non-seedling sample data, enumerate candidate band combinations in historical hyperspectral image data and calculate the band separability score data for each candidate band combination. The band separability score data is determined by the inter-class difference and the intra-class dispersion.

[0030] Candidate band combinations are enumerated from the available band set of historical hyperspectral image data, and the band separability score data for each candidate band combination is calculated. In the embodiment, for any candidate band combination, the reflectance of its corresponding pixels is concatenated to form pixel vegetation feature vector data; the feature mean vector and covariance matrix of seedling sample data and the feature mean vector and covariance matrix of non-seedling sample data are calculated respectively, and the inter-class variability and intra-class variability are defined accordingly. The inter-class difference is taken as the square of the L2 norm of the difference between the two class mean vectors, and the intra-class dispersion is taken as the sum of the traces of the two class covariance matrices. The band separability score data can be taken in Fisher form, that is, the ratio of the inter-class difference to the intra-class dispersion, where the mean vector is obtained by averaging the sample vectors dimension by dimension, the covariance matrix is ​​obtained by averaging the outer products of the sample vectors after removing the mean, and the trace of the covariance matrix is ​​the element on the diagonal of the matrix. The larger the band separability score data, the stronger the distinction between seedlings and non-seedlings of the candidate band combination.

[0031] Step S103: Select the candidate band combination with the largest band separability score data as the planting discrimination band group and store it.

[0032] The candidate band combination with the largest band separability score is selected as the vegetation discrimination band group and stored. To ensure cross-seasonal robustness, this embodiment calculates scores for different historical dates and takes a weighted average score. The weights are determined by the image quality score (cloud cover and blur). The final output vegetation discrimination band group is used as the fixed input channel set for the subsequent vegetation confidence map.

[0033] Step S2 includes the following sub-steps: Step S201: Perform radiometric normalization processing on periodic hyperspectral image data and flight attitude positioning data to output periodic reflectance image data.

[0034] Periodic hyperspectral image data, periodic RGB image data, and flight attitude positioning data are acquired. The flight attitude positioning data includes at least: timestamp for each frame, latitude and longitude, elevation, heading angle, pitch angle, roll angle, and camera intrinsic parameters. Radiometric normalization is performed on the periodic hyperspectral image data to output periodic reflectance image data. In this embodiment, the original grayscale value of a pixel in a certain band is used as the original observation value, the dark field calibration grayscale value as the dark field value, and the white board calibration grayscale value as the white board value. The reflectance of that band is then the ratio of the difference between the original observation value and the dark field value to the difference between the white board value and the dark field value. Periodic reflectance image data is calculated pixel-by-pixel for all bands. If there are differences in solar altitude angle, this embodiment further compensates for the reflectance by the incident angle, i.e., dividing the reflectance by the cosine of the incident term and truncating it to the [0,1] interval. The compensation parameter is calculated from the flight time and geographical location.

[0035] Step S202: Based on the flight attitude positioning data, orthorectify the periodic reflectivity image data, and perform feature matching registration between periodic reflectivity image data of adjacent periods to obtain consistent spatial coordinates across periods.

[0036] Orthorectification is performed on periodic reflectance image data based on flight attitude positioning data, and feature matching registration is performed between adjacent periods to obtain consistent spatial coordinates across periods. In this embodiment, orthorectification first projects pixels onto a unified geographic coordinate system using flight attitude positioning data and a terrain elevation model; then, stable feature points (field ridge intersections, road corners, ditch inflection points) are extracted from periodic RGB image data, and corresponding point pairs are obtained through feature descriptor matching between adjacent periods. A two-dimensional affine transformation or homography transformation is estimated using random sampling consistency estimation; this transformation is applied to the periodic reflectance image data to ensure spatial alignment of reflectance image data from each period on the same pixel grid. Registration quality is measured by the proportion of matched inliers and reprojection error, and stored as a quality field in the evidence package data.

[0037] Step S203: Align the registered periodic reflectance image data with the periodic RGB image data in terms of coordinates, and output the reflectance aligned dataset.

[0038] The registered periodic reflectance image data is coordinate-aligned with the periodic RGB image data to output a reflectance-aligned dataset. The reflectance-aligned dataset contains at least: aligned periodic reflectance image data, aligned periodic RGB image data, geographic coordinate index of each pixel, registration quality score, and shadow mask index. The shadow mask is generated jointly by RGB brightness threshold and local contrast and is used to reduce the weight of shadow areas in subsequent confidence calculations.

[0039] The vegetation confidence map is generated in step S3 as follows: Pixel vegetation feature vector data is constructed based on the pixel reflectance data corresponding to the vegetation discrimination band group in the reflectance aligned dataset.

[0040] Based on the statistical analysis of seedling sample data, seedling spectral template data was obtained, and the spectral similarity data between pixel vegetation feature vector data and seedling spectral template data was calculated.

[0041] The spectral similarity data are normalized and output as a vegetation confidence map.

[0042] Pixel vegetation feature vector data is constructed based on pixel reflectance data corresponding to the vegetation discrimination band group in the reflectance aligned dataset, and seedling spectral template data is formed using historical samples. In this embodiment, the seedling spectral template data is taken as the mean vector of the seedling sample data on the vegetation discrimination band group; for any pixel, the spectral similarity data between its pixel vegetation feature vector and the seedling spectral template is calculated. The spectral similarity can be taken as cosine similarity, which is the ratio of the dot product of the pixel vector and the template vector to the product of their norms; then the similarity is linearly normalized to [0,1] to obtain the vegetation confidence map. To enhance the suppression of mulch film reflection and wet soil bright spots, this embodiment also calculates the Mahalanobis distance from the pixel vector to the template vector and uses an exponential function to form a penalty term. Finally, the vegetation confidence map is taken as the product of the normalized cosine similarity value and the penalty term. The smaller the penalty term, the greater the spectral deviation.

[0043] In step S2, the flight attitude positioning data is used as a geometric registration parameter, rather than as an image pixel value for fusion. Specifically, for any frame in the periodic hyperspectral image data, the flight attitude positioning data of that frame includes the position vector of the camera center in the geographic coordinate system, the attitude rotation matrix, and the camera intrinsic parameter matrix. For the pixel coordinates of any pixel in that frame, the pixel coordinates are first back-projected into the line of sight in the camera coordinate system by the camera intrinsic parameter matrix, and then the line of sight is transformed into the geographic coordinate system by the attitude rotation matrix to obtain the ground ray corresponding to that pixel. Then, the intersection with the terrain elevation model is obtained to obtain the geographic coordinates of that pixel, thereby forming geographic coordinate mapping data from pixel coordinates to geographic coordinates. Based on this, the periodic hyperspectral image data is resampled to a unified geographic grid to obtain periodic reflectance image data. Between adjacent periods, stable feature points such as road inflection points, field ridge intersections, and ditch inflection points are extracted and matched from the periodic RGB image data. The cross-period two-dimensional geometric transformation is estimated and applied to the periodic reflectance image data to align the periodic reflectance image data of each period on the same geographic grid. Finally, the registered periodic reflectance image data, periodic RGB image data, and their geographic coordinate mapping data are encapsulated together into a reflectance aligned dataset.

[0044] The vegetation confidence map is used to quantify the matching degree of each pixel in the reflectance aligned dataset to the seedling vegetation, so that the extraction of subsequent candidate seedling center point data changes from bright spots to pixel peaks that conform to the seedling spectrum, thereby suppressing non-vegetation bright interference such as mulch reflection and wet soil bright spots. At the same time, the value of the vegetation confidence map is used as the stopping condition for regional growth to prevent candidate patches from expanding to soil or shadow boundaries. In addition, in cross-period association, the confidence of the center point trajectory over time is used as part of the growth consistency to participate in the calculation of temporal growth probability, which is used to distinguish between short-term noise points and continuous growth targets.

[0045] Step S3 includes the following sub-steps: Step S301: Detect the location of local maxima in the vegetation confidence map and output the initial center point data.

[0046] The system detects local maxima locations in the vegetation confidence map and outputs initial center point data. In this embodiment, a sliding window non-maximum suppression is used: within a window centered on a pixel, if the confidence of that pixel is the maximum of the window and exceeds an adaptive threshold, it is retained as the initial center point; the adaptive threshold is taken as the upper quantile of the built-in confidence distribution of the land parcel registration boundary data to ensure that the number of candidate points can be stably output under different lighting conditions.

[0047] Step S302: Using the initial center point data as seeds, perform region growing in the reflectance aligned dataset to obtain candidate patch data, and calculate the patch spectral consistency data and patch shape feature data corresponding to the candidate patch data.

[0048] Using the initial center point data as seeds, region growing is performed on the reflectance-aligned dataset to obtain candidate patch data, and patch spectral consistency data and patch shape feature data are calculated. In this embodiment, region growing is expanded using a dual-threshold method with a confidence difference threshold and a reflectance difference threshold: For adjacent pixels of the initial center point, if the absolute value of the difference between its confidence score and the center point confidence score is less than the threshold and the Euclidean distance between its reflectance vector on the plant discrimination band group and the center point vector is less than the threshold, then it is included in the patch. The patch pattern consistency data is taken as the average similarity between the pixel vector within the patch and the seedling pattern template; the patch shape feature data includes at least the patch area, perimeter, roundness (calculated as the square of the ratio of 4π times the area to the perimeter) and elongation ratio (calculated by the ratio of the second moment principal axes of the patch), which are used to exclude non-seedling structures such as linear field ridges and edge shadow stripes.

[0049] Step S303: Based on the patch pattern consistency data and patch shape feature data, remove non-seedling candidate patch data, and output the centroid of the remaining candidate patch data as the candidate seedling center point data.

[0050] Based on patch pattern consistency data and patch shape feature data, non-seedling candidate patch data are eliminated, and candidate seedling center point data are output. In this embodiment, the elimination logic is as follows: When the patch pattern consistency data is below the threshold, the patch roundness is below the threshold, or the patch area exceeds the reasonable range of the seedling area, the candidate patch is determined to be a non-seedling candidate patch and is removed; the centroid of the remaining candidate patches is used as the candidate seedling center point data, where the centroid coordinates are the average of the pixel coordinates within the patch, and its geographic coordinate index is inherited to support subsequent cross-period association.

[0051] Step S4 includes the following sub-steps: Step S401: Perform directional statistics on the center point data of candidate seedlings and fit the data to obtain the main sowing direction data.

[0052] The direction statistics of the candidate seedling center point data are statistically analyzed and fitted to obtain the sowing main direction data. In the example, within the registered boundary data of each plot, the set of nearest neighbor line directions of the candidate seedling center point data is taken, the direction angle is statistically analyzed by histogram, and the main peak direction is selected as the sowing main direction data; when there are two main peaks (staggered row sowing), the main peak and the secondary peak are taken, and the maximum response is taken in the subsequent probability calculation.

[0053] Step S402: Statistically analyze the spacing distribution of adjacent candidate seedling center point data along the main sowing direction, and output the main spacing data.

[0054] The spacing distribution of adjacent candidate seedling center points along the main sowing direction is statistically analyzed, and the main spacing data is output. In this embodiment, the center point coordinates are projected onto the direction perpendicular to the main sowing direction, and the projection difference is clustered or peak detection is performed to obtain the row spacing main spacing; simultaneously, peak detection is performed on the distance between adjacent points along the main sowing direction to obtain the plant spacing main spacing. To avoid pixel scale inconsistencies, this embodiment uses the reflectance-aligned dataset's geographic coordinate index to uniformly convert the distances to meters.

[0055] Step S403: Calculate spatial pattern probability data based on the deviation of the candidate seedling center point data from the sowing main direction data and the spacing deviation from the main spacing data.

[0056] Spatial pattern probability data is calculated based on deviation and spacing deviation. In this embodiment, for any candidate seedling center point, the directional deviation of its local neighborhood point set relative to the main sowing direction is calculated (the absolute value of the angle between the local fitting direction and the main sowing direction), the distance deviation from the nearest row is calculated (the absolute value of the difference between the distance to the center line of the nearest row and the main spacing between rows), and the plant spacing deviation from the nearest point is calculated (the absolute value of the difference between the nearest neighbor distance and the main spacing between plants). These three values ​​are then mapped to [0,1] using a Gaussian function and multiplied together to obtain the spatial pattern probability data. The scale parameter in the above mapping is adaptively determined by the median and interquartile range of the deviation distribution within the plot, thus remaining stable under different plots and different seedling densities.

[0057] Spatial regularity probability data is used to characterize whether candidate seedling center point data present a consistent row orientation and approximately constant row and plant spacing in the tillage spatial structure within the plot. Thus, in the early seedling stage or when some periods are missing data, tillage geometry can still be used to distinguish randomly distributed weeds or noise points from regularly sown seedling points, and this data can be used as a spatial prior for the comprehensive tillage confidence score in the fusion process.

[0058] The candidate seedling center point data, as the centroid coordinates of candidate patches, is the minimum stable geometric descriptor for cross-period association. The calculation of temporal growth probability uses the center point trajectory as the carrier, and identifies targets that appear continuously in the vicinity of the same geographical location in multiple periods and whose candidate patch area and confidence show an increasing trend as real seedling growth, thereby avoiding misjudgment caused by relying solely on candidate points in a single period.

[0059] Step S5 includes the following sub-steps: Step S501: The candidate seedling center point data for each period are stored in partitions according to the land parcel registration boundary data to form a time-series center point dataset.

[0060] The candidate seedling center point data for each period are partitioned and stored according to the land parcel registration boundary data to form a time-series center point dataset. In the embodiment, the geographic coordinate index of the reflectance-aligned dataset is used to ensure that the center point of each period falls within the corresponding land parcel registration boundary data, and the period identifier, center point coordinates, corresponding candidate patch index, vegetation confidence value, and patch shape feature data are written into the database to form a time-series center point dataset organized by land parcel.

[0061] Step S502: Perform cross-cycle correlation based on the time-series center point dataset, output center point trajectory data, and calculate growth consistency data based on the candidate patch data corresponding to the center point trajectory data.

[0062] Based on the time-series center point dataset, cross-period correlation is performed to output center point trajectory data, and growth consistency data is calculated based on the candidate patch data corresponding to the center point trajectory data. In this embodiment, cross-period correlation employs geographic distance gating and minimum cost matching. For each center point in the current period, candidate matches with a geographical distance less than the gating radius are searched in the set of center points in the previous period. The gating radius is determined by the ground resolution and the upper bound of the registration error. If there are multiple candidate matches, the one with the smallest weighted sum of distance cost, spectral difference cost, and shape difference cost will be the match. The point sequence obtained by continuous matching is defined as the center point trajectory data. The growth consistency data is composed of the increment of candidate patch area on the trajectory over time, the increment of vegetation confidence value over time, and the stability of spectral consistency over time. The area increment can be taken as the area difference between adjacent periods, and the stability can be taken as the variance of spectral consistency on the trajectory and reverse normalized.

[0063] Step S503: When a certain period is missing, interpolation is performed based on the center point trajectory data and growth consistency data of adjacent periods, and the confidence weight is reduced to output the time series growth probability data.

[0064] When a certain period is missing a measurement, interpolation is performed and the confidence weight is reduced to output temporal growth probability data. In the embodiment, missing measurement refers to the situation where the quality score of the reflectance aligned dataset for that period is lower than the threshold or the shadow mask coverage is higher than the threshold, making the center point unusable. In this case, the center point trajectory data is linearly extrapolated to predict the center point position and patch area for that period, and the growth consistency contribution for that period is multiplied by the missing measurement attenuation coefficient. Finally, for each center point trajectory, the temporal growth probability data is obtained by weighted averaging the growth consistency of each period within the window, and the window length is set to the K most recent periods to adapt to situations such as patrol grounding.

[0065] Step S504: The spatial pattern probability data and the temporal growth probability data are fused according to weights to output comprehensive farming confidence data.

[0066] Spatial pattern probability data and temporal growth probability data are fused together according to weights to output comprehensive cultivation confidence data. In this example, the comprehensive cultivation confidence data is a weighted sum. In the early seedling stage (determined by the small average area of ​​patches within the plot), the weight of spatial pattern probability data is increased, and in the stable growth stage, the weight of temporal growth probability data is increased. The weights are adaptively updated with the cycle, so that the early stage relies on the row sowing pattern and the later stage relies on the consistency of continuous growth, thereby reducing misjudgments caused by weeds and short-term light fluctuations.

[0067] Step S6 includes the following sub-steps: Step S601: The comprehensive cultivation confidence data is rasterized within the land parcel registration boundary data and connected components are extracted to obtain cultivation connected component data.

[0068] The comprehensive cultivation confidence data is rasterized within the plot registration boundary data, and connected component extraction is performed to obtain cultivated connected component data. In this embodiment, the plot registration boundary data defines the raster range, and the comprehensive cultivation confidence of each pixel is calculated (obtained by interpolation of the confidence of the center point near the pixel or directly calculated from the pixel). Then, the raster is binarized with an adaptive threshold: the threshold is taken as a function between the upper and lower quantiles of the plot's built-in confidence distribution to adapt to different crop density; connected component labeling is performed on the binary raster to obtain cultivated connected component data, and the area and perimeter of each connected component are retained.

[0069] Step S602: Extract the concave contour from the cultivated connected domain data and output the cultivated boundary polygon data.

[0070] The concave contour of the cultivated connected component data is extracted to output cultivated boundary polygon data. In the embodiment, the boundary point set of each connected component is extracted and a concave contour is generated using the α-shape or concave shell algorithm to obtain cultivated boundary polygon data that fits the actual cultivated shape of the plot; at the same time, the polygons are smoothed and small holes are filled. The smoothing scale is determined by the ground resolution to ensure that the output boundary is both fitting and unaffected by noise and jagged edges.

[0071] Step S603: Overlay the cultivated boundary polygon data with the land registration boundary data to output the over-boundary occupation polygon data, and write the cultivated boundary polygon data and the over-boundary occupation polygon data into the evidence package data.

[0072] The cultivated boundary polygon data is overlaid with the plot registration boundary data to output the out-of-bounds occupied polygon data, which is then written into the evidence package data. In this embodiment, the out-of-bounds occupied polygon data is defined as the difference set of the cultivated boundary polygon data outside the plot registration boundary data, and its area is calculated simultaneously. The evidence package data includes at least: periodic RGB image data slices, reflectance aligned dataset index, vegetation confidence map index, cultivated boundary polygon data, out-of-bounds occupied polygon data, registration quality score, shadow mask coverage, and key thresholds.

[0073] The logic for generating status result data is as follows: Cultivated coverage data is calculated based on cultivated boundary polygon data and plot registration boundary data; the proportion of land occupied beyond the boundary is calculated based on the polygon data of land occupied beyond the boundary; and the risk of land abandonment is calculated based on time-series growth probability data.

[0074] When the cultivated coverage data meets the first threshold and the fallow risk data meets the second threshold, the normal cultivated status is output.

[0075] When the cultivated coverage data does not meet the first threshold and the fallow risk data does not meet the second threshold, a suspected fallow status is output.

[0076] When the out-of-bounds occupancy rate data meets the third threshold, a suspected illegal occupancy status is output.

[0077] The status result data, along with the corresponding periodic RGB image data slices, reflectance aligned dataset index, cultivated boundary polygon data, and out-of-bounds occupied polygon data, are uniformly packaged and output as evidence package data.

[0078] Cultivated land coverage data is calculated based on cultivated boundary polygon data and plot registration boundary data. Over-boundary occupation ratio data is calculated based on over-boundary occupation polygon data. Abandonment risk data is calculated based on time-series growth probability data, and the status result data is output. Example: In the same coordinate system, the cultivated boundary polygon data and the land registration boundary data are overlaid. First, the area of ​​the overlapping area is calculated. Then, the area is divided by the area of ​​the land registration boundary data. The resulting ratio is the cultivated coverage rate data. Overlay the cultivated boundary polygon data with the land parcel registration boundary data, take the area of ​​the cultivated boundary polygon data that exceeds the land parcel registration boundary data as the over-boundary occupation polygon data, calculate the area of ​​the over-boundary area, and then divide it by the area of ​​the land parcel registration boundary data to obtain the over-boundary occupation ratio data. Within each plot, the temporal growth probability data corresponding to all center point trajectory data are averaged to obtain the overall growth continuity level of the plot. Then, the average value is subtracted from 1 to obtain the abandonment risk data. If there are missing measurement periods, the weight of trajectories with more missing measurements is reduced so that their contribution to the average value is smaller, thereby avoiding misjudgment of abandonment due to flight stoppages and rainy weather.

[0079] For missing measurement periods, a decay weight is applied. When the cultivated coverage rate meets the first threshold and the fallow risk data meets the second threshold, the normal cultivated status is output. When the cultivated coverage rate does not meet the first threshold and the fallow risk data does not meet the second threshold, the suspected fallow status is output. When the proportion of over-boundary occupation meets the third threshold, the suspected illegal occupation status is output. The status result data and evidence package data are output together to support the inspection, verification and law enforcement evidence collection for cultivated land protection.

[0080] Example 2, refer to Figure 2 This paper provides a farmland status monitoring system for farmland protection, including a data retrieval module, a data alignment module, a center point extraction module, a pattern calculation module, a cultivation determination module, and an evidence output module.

[0081] The data retrieval module is used to retrieve historical hyperspectral image data, historical labeled sample data, and land parcel registration boundary data to determine the vegetation discrimination band group.

[0082] The data alignment module is used to collect periodic hyperspectral image data, periodic RGB image data and flight attitude positioning data. Based on the flight attitude positioning data, orthorectification and cross-period registration are performed to generate a reflectance aligned dataset containing geographic coordinate mapping data.

[0083] The center point extraction module is used to extract pixel reflectance from the reflectance aligned dataset based on the vegetation discrimination band group and calculate the vegetation confidence map. It detects local maxima and performs region growth to obtain candidate patch data. The centroid of the candidate patch data is output as candidate seedling center point data.

[0084] The pattern calculation module is used to calculate spatial pattern probability data based on the candidate seedling center point data.

[0085] The cultivation determination module is used to associate multi-period center points to form center point trajectory data and calculate temporal growth probability data. It integrates spatial regularity probability data with temporal growth probability data to obtain comprehensive cultivation confidence data.

[0086] The evidence output module is used to extract cultivated boundary polygon data based on comprehensive cultivated confidence data, compare it with the land parcel registration boundary data, and output status result data and evidence package data.

[0087] This invention constructs a vegetation confidence map using vegetation discrimination band groups and combines spectral consistency and patch shape characteristics to eliminate non-seedling candidates, making it difficult to misidentify mulch film reflection, wet soil bright spots, field ridge linear structures, shadow boundaries, etc. as cultivation targets, thereby improving the accuracy and stability of cultivation identification.

[0088] A reflectance-aligned dataset is generated by radiometric normalization, orthorectification, and feature matching registration, so that images from different periods are in a unified coordinate system. On this basis, the center point trajectory is established and growth consistency is calculated to avoid misjudgment that looks like vegetation due to a single time-phase threshold, thereby enhancing the credibility of fallow and reseeding verification.

[0089] By introducing spatial regularity probability data and using the main sowing direction and main spacing to constrain the spatial arrangement of candidate seedling center points, a high differentiation ability can still be obtained based on the row sowing structure in the early stage of seedling emergence (when vegetation cover is low and texture is weak), thus reducing early missed detections.

[0090] The system employs a multi-period window to calculate the time series growth probability and performs interpolation and weight decay processing on missing periods. This ensures continuous judgment even when there are rainy days, no-fly zones, cloud cover, or poor quality in individual periods, reducing the risk of misjudgment and abandonment.

[0091] By rasterizing the comprehensive cultivation confidence score, extracting connected components, and generating concave contours, we obtain the cultivation boundary polygon data, which is then overlaid with the land parcel registration boundary data to obtain the over-boundary occupation polygon data. At the same time, we encapsulate image slices, alignment indexes, boundary results, and quality scores to form evidence package data, which facilitates regulatory review, rectification verification, and administrative law enforcement evidence collection.

[0092] Based on data on cultivated land coverage, the proportion of land encroachment, and the risk of land abandonment, the output status results can be directly used for county-level ledger management, risk classification, rectification assignment and review, thereby improving the efficiency of farmland protection inspections and handling.

[0093] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Read-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0094] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the protection scope of the present invention.

Claims

1. A method for monitoring the status of arable land for arable land protection, characterized in that, Includes the following steps: Step S1: Retrieve historical hyperspectral image data, historical labeled sample data, and land parcel registration boundary data to determine the vegetation discrimination band group; Step S2: Collect periodic hyperspectral image data, periodic RGB image data and flight attitude positioning data. Perform orthorectification and cross-period registration based on the flight attitude positioning data to generate a reflectance aligned dataset containing geographic coordinate mapping data. Step S3: Extract pixel reflectance from reflectance aligned dataset based on vegetation discrimination band group and calculate vegetation confidence map, detect local maxima and obtain candidate patch data by region growth, and output the centroid of candidate patch data as candidate seedling center point data. Step S4: Calculate spatial pattern probability data based on the candidate seedling center point data; Step S5: Correlate the center points of multiple cycles to form center point trajectory data and calculate the temporal growth probability data. Then, merge the spatial regularity probability data and the temporal growth probability data to obtain comprehensive farming confidence data. Step S6: Extract the cultivated boundary polygon data based on the comprehensive cultivated confidence data, compare it with the land parcel registration boundary data, and output the status result data and evidence package data.

2. The method for monitoring the status of arable land for arable land protection as described in claim 1, characterized in that, Step S1 includes the following sub-steps: Step S101: Retrieve historical hyperspectral image data and historical labeled sample data from the database. The historical labeled sample data includes seedling sample data and non-seedling sample data. Step S102: Based on seedling sample data and non-seedling sample data, enumerate candidate band combinations in historical hyperspectral image data and calculate the band separability score data of each candidate band combination. The band separability score data is determined by the inter-class difference and the intra-class dispersion. Step S103: Select the candidate band combination with the largest band separability score data as the planting discrimination band group and store it.

3. The method for monitoring the status of arable land for arable land protection as described in claim 2, characterized in that, Step S2 includes the following sub-steps: Step S201: Perform radiometric normalization processing on periodic hyperspectral image data and flight attitude positioning data to output periodic reflectance image data; Step S202: Orthorectify the periodic reflectivity image data based on the flight attitude positioning data, and perform feature matching and registration between periodic reflectivity image data of adjacent periods to obtain consistent spatial coordinates across periods. Step S203: Align the registered periodic reflectance image data with the periodic RGB image data in terms of coordinates, and output the reflectance aligned dataset.

4. The method for monitoring the status of arable land for arable land protection as described in claim 3, characterized in that, The vegetation confidence map is generated in step S3 as follows: Pixel vegetation feature vector data is constructed based on the pixel reflectance data corresponding to the vegetation discrimination band group in the reflectance alignment dataset. Based on the statistical analysis of seedling sample data, seedling pattern template data was obtained, and the pattern similarity data between pixel vegetation feature vector data and seedling pattern template data was calculated. The spectral similarity data are normalized and output as a vegetation confidence map.

5. A method for monitoring the status of arable land for arable land protection as described in claim 4, characterized in that, Step S3 includes the following sub-steps: Step S301: Detect the location of local maxima in the vegetation confidence map and output the initial center point data; Step S302: Using the initial center point data as seeds, perform region growth in the reflectance aligned dataset to obtain candidate patch data, and calculate the patch spectral consistency data and patch shape feature data corresponding to the candidate patch data; Step S303: Based on the patch pattern consistency data and patch shape feature data, remove non-seedling candidate patch data, and output the centroid of the remaining candidate patch data as the candidate seedling center point data.

6. A method for monitoring the status of arable land for arable land protection as described in claim 5, characterized in that, Step S4 includes the following sub-steps: Step S401: Perform directional statistics on the center point data of candidate seedlings and fit the data to obtain the main sowing direction data; Step S402: Statistically analyze the spacing distribution of adjacent candidate seedling center point data along the main sowing direction, and output the main spacing data; Step S403: Calculate spatial pattern probability data based on the deviation of the candidate seedling center point data from the sowing main direction data and the spacing deviation from the main spacing data.

7. A method for monitoring the status of arable land for arable land protection as described in claim 6, characterized in that, Step S5 includes the following sub-steps: Step S501: The candidate seedling center point data of each period are stored in partitions according to the land parcel registration boundary data to form a time-series center point dataset; Step S502: Perform cross-cycle correlation based on the time series center point dataset, output center point trajectory data, and calculate growth consistency data based on the candidate patch data corresponding to the center point trajectory data; Step S503: When a certain period is missing, interpolation is performed based on the center point trajectory data and growth consistency data of adjacent periods, and the confidence weight is reduced to output the time series growth probability data. Step S504: The spatial pattern probability data and the temporal growth probability data are fused according to weights to output comprehensive farming confidence data.

8. A method for monitoring the status of arable land for arable land protection as described in claim 7, characterized in that, Step S6 includes the following sub-steps: Step S601: Rasterize the comprehensive cultivation confidence data within the land parcel registration boundary data and extract the connected components to obtain the cultivation connected component data; Step S602: Extract the concave contour from the cultivated connected component data and output the cultivated boundary polygon data; Step S603: Overlay the cultivated boundary polygon data with the land registration boundary data to output the over-boundary occupation polygon data, and write the cultivated boundary polygon data and the over-boundary occupation polygon data into the evidence package data.

9. A method for monitoring the status of arable land for arable land protection as described in claim 8, characterized in that, The logic for generating the status result data is as follows: Cultivated coverage data is calculated based on cultivated boundary polygon data and plot registration boundary data; the proportion of land occupation is calculated based on land occupation polygon data; and the risk of land abandonment is calculated based on time-series growth probability data. When the cultivated coverage data meets the first threshold and the fallow risk data meets the second threshold, the normal cultivated status is output. When the cultivated coverage data does not meet the first threshold and the fallow risk data does not meet the second threshold, output a suspected fallow status. When the out-of-bounds occupancy rate data meets the third threshold, a suspected illegal occupancy status is output. The status result data, along with the corresponding periodic RGB image data slices, reflectance aligned dataset index, cultivated boundary polygon data, and out-of-bounds occupied polygon data, are uniformly packaged and output as evidence package data.

10. A farmland status monitoring system for farmland protection, which is applied in a farmland status monitoring method for farmland protection as described in any one of claims 1-9, characterized in that, It includes a data retrieval module, a data alignment module, a center point extraction module, a pattern calculation module, a cultivation determination module, and an evidence output module; The data retrieval module is used to retrieve historical hyperspectral image data, historical labeled sample data, and land parcel registration boundary data to determine the vegetation discrimination band group; The data alignment module is used to collect periodic hyperspectral image data, periodic RGB image data and flight attitude positioning data, perform orthorectification and cross-period registration based on the flight attitude positioning data, and generate a reflectance aligned dataset containing geographic coordinate mapping data. The center point extraction module is used to extract pixel reflectance from reflectance aligned dataset based on vegetation discrimination band group and calculate vegetation confidence map, detect local maxima and obtain candidate patch data by region growth, and output the centroid of candidate patch data as candidate seedling center point data. The pattern calculation module is used to calculate spatial pattern probability data based on the candidate seedling center point data; The cultivation determination module is used to associate multi-period center points to form center point trajectory data and calculate temporal growth probability data, and to integrate spatial regularity probability data with temporal growth probability data to obtain comprehensive cultivation confidence data. The evidence output module is used to extract cultivated boundary polygon data based on comprehensive cultivated confidence data, compare it with the land parcel registration boundary data, and output status result data and evidence package data.

Images