Method and system for dynamically monitoring cultivated land and grain planting conditions and storage medium

Abstract

The invention relates to the technical field of intelligent monitoring, and discloses a farmland grain planting condition dynamic monitoring method and system and a storage medium, and the method comprises the steps: obtaining a multi-temporal optical remote sensing image of a target area, and reconstructing the multi-temporal optical remote sensing image to obtain a high-resolution farmland image; performing feature extraction and crop identification on the high-resolution farmland image to obtain a crop type distribution diagram; and performing multi-time comparative analysis on the crop type distribution diagram to obtain cultivated land utilization change data, and generating a cultivated land grain planting condition monitoring report based on the cultivated land utilization change data. Precise distinguishing between grain crops such as rice, wheat and corn and commercial crops is achieved, fine non-grain changes are effectively recognized, and the accuracy of dynamic monitoring of cultivated land and grain planting conditions is improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent monitoring technology, and in particular to a method, system and storage medium for dynamic monitoring of grain planting conditions on arable land. Background Technology

[0002] Traditional methods for monitoring arable land rely on manual field surveys and simple remote sensing image analysis, which suffer from problems such as long monitoring cycles, limited coverage, and high labor costs, making it difficult to meet the requirements of large-scale, high-frequency dynamic monitoring. Furthermore, existing remote sensing monitoring technologies do not fully consider the specific characteristics of farmland scenarios when dealing with complex agricultural landscapes. They lack targeted solutions for quality degradation during satellite signal transmission and ignore the specific requirements of crop identification accuracy for agricultural monitoring tasks. As a result, although the reconstructed images are visually improved, they still have shortcomings in crop type differentiation and boundary recognition, affecting the accuracy of crop identification and arable land change detection. Summary of the Invention

[0003] This invention provides a method, system, and storage medium for dynamic monitoring of grain cultivation in arable land. This invention effectively solves the problem of quality degradation of satellite remote sensing images during transmission, enables accurate differentiation between grain crops such as rice, wheat, and corn and cash crops, effectively identifies subtle non-grain changes, and improves the accuracy of dynamic monitoring of grain cultivation in arable land.

[0004] In a first aspect, the present invention provides a method for dynamic monitoring of grain cultivation on arable land, the method comprising: Acquire multi-temporal optical remote sensing images of the target area and reconstruct the multi-temporal optical remote sensing images to obtain high-resolution farmland images; Feature extraction and crop identification are performed on the high-resolution farmland images to obtain a crop type distribution map; A multi-time comparative analysis was performed on the crop type distribution map to obtain farmland use change data, and a farmland grain planting monitoring report was generated based on the farmland use change data.

[0005] In conjunction with the first aspect, in a first implementation of the first aspect of the present invention, the step of acquiring multi-temporal optical remote sensing images of the target area and reconstructing the multi-temporal optical remote sensing images to obtain high-resolution farmland images includes: Acquire multi-temporal optical remote sensing images of the target area; Analyze the spatial basis functions of farmland plots in the multi-temporal optical remote sensing images, and combine the spatial basis functions to form a sparse basis dictionary for farmland. The texture response patterns of different crop types in each spectral band in the multi-temporal optical remote sensing images were statistically analyzed, and a farmland texture correlation matrix was constructed based on the texture response patterns. The temporal information of the multi-temporal optical remote sensing images is acquired and the corresponding crop growth period is identified. The crop phenological weight factor is calculated based on the crop growth period, and the sparsity threshold is determined using the crop phenological weight factor. Based on the farmland sparse base dictionary, the farmland texture correlation matrix, and the sparsity threshold, the transmission channel of the multi-temporal optical remote sensing image is iteratively reconstructed to obtain the channel response matrix; Based on the channel response matrix, dual-drive super-resolution reconstruction is performed on the multi-temporal optical remote sensing images to obtain high-resolution farmland images.

[0006] In conjunction with the first aspect, in a second implementation of the first aspect of the present invention, the step of performing dual-drive super-resolution reconstruction of the multi-temporal optical remote sensing image based on the channel response matrix to obtain a high-resolution farmland image includes: A sparse reconstruction model is established using the farmland sparse basis dictionary and the farmland texture correlation matrix, and the transmission channel of the multi-temporal optical remote sensing image is represented as the initial sparse processing result of sparse basis functions and sparse coefficients. Candidate basis functions are selected one by one from the spatial basis functions of the farmland sparse basis dictionary. The similarity between the candidate basis functions and the known farmland plot shapes is calculated. Basis functions with similarity greater than a preset threshold are selected as structural constraint basis functions. Based on the sparse reconstruction model and the structural constraint basis function, the initial sparse processing result is reconstructed to generate the reconstruction result; Based on the reconstruction results and the characteristic spectral curves, the candidate basis function with the smallest spectral angular distance is selected as the optimal basis function, and the residual values ​​are updated. The number of iterations is controlled according to the sparsity threshold. The method is to repeatedly select the method that minimizes the residual value and satisfies the optimal basis function until the residual converges or reaches the sparsity threshold, thereby obtaining the channel response matrix.

[0007] In conjunction with the first aspect, in a third implementation of the first aspect of the present invention, the step of performing dual-drive super-resolution reconstruction of the multi-temporal optical remote sensing image based on the channel response matrix to obtain a high-resolution farmland image includes: The channel response matrix and the multi-temporal optical remote sensing image are input into a dual-drive super-resolution reconstruction network. Multi-scale feature extraction is performed through the encoder in the dual-drive super-resolution reconstruction network to obtain a coded feature map. Based on the encoded feature map, a boundary preservation driving process is performed to generate a boundary enhancement feature map, and the boundary enhancement feature map is then subjected to spectral feature enhancement to obtain a spectral enhancement feature map. The spectral enhancement feature map is upsampled and reconstructed by the decoder in the dual-drive super-resolution reconstruction network, while maintaining the plot boundary clarity of the boundary enhancement feature map and the crop differentiation features of the spectral enhancement feature map, to obtain a high-resolution farmland image.

[0008] In conjunction with the first aspect, in the fourth implementation of the first aspect of the present invention, the step of inputting the channel response matrix and the multi-temporal optical remote sensing image into a dual-drive super-resolution reconstruction network, and performing multi-scale feature extraction through the encoder in the dual-drive super-resolution reconstruction network to obtain an encoded feature map includes: The channel response matrix is ​​used to compensate for transmission distortion in the multi-temporal optical remote sensing image to obtain a channel-corrected image. The channel-corrected image is input into the feature extraction module of the encoder in the dual-drive super-resolution reconstruction network for feature extraction to obtain the original features at each scale. Channel attention processing is performed on the original features at each scale to obtain channel-enhanced features; The channel enhancement features are adaptively adjusted based on the quality assessment information of the channel response matrix, and the encoded feature map is output.

[0009] In conjunction with the first aspect, in the fifth implementation of the first aspect of the present invention, the step of extracting features and identifying crops from the high-resolution farmland image to obtain a crop type distribution map includes: Scale matching features were obtained by performing area statistical analysis on farmland plots in the high-resolution farmland image. The scale-matching features are input into a multi-scale feature pyramid for parallel processing to obtain multi-scale spatial features. Based on the time series information of the multi-temporal optical remote sensing images, the multi-scale spatial features are input into the temporal attention module for temporal dimension fusion to obtain temporal enhancement features; Based on the aforementioned temporal enhancement features, feature correction is performed by combining the geometric boundary information of farmland plots and crop phenological characteristics to obtain a multi-scale feature map of farmland. Crop identification is performed based on the multi-scale feature map of the farmland to obtain a crop type distribution map.

[0010] In conjunction with the first aspect, in the sixth implementation of the first aspect of the present invention, the step of identifying crops based on the multi-scale feature map of the farmland to obtain a crop type distribution map includes: The multi-scale feature map of the farmland is input into the spectral feature separation layer of the crop recognition model to separate spectral components and obtain the spectral separation result. Based on the grain crop components in the spectral separation results, crop feature enhancement is performed to obtain enhanced spectral features; The enhanced spectral features are input into the crop classification layer of the crop recognition model for type identification to obtain crop classification results. Based on the crop classification results, a connected region analysis is performed, and a crop type distribution map is generated by combining farmland plot boundary information.

[0011] In conjunction with the first aspect, in the seventh implementation of the first aspect of the present invention, the step of performing multi-time comparative analysis on the crop type distribution map to obtain arable land use change data, and generating a arable land grain planting monitoring report based on the arable land use change data, includes: The distribution maps of the crop types at different time points are compared pixel by pixel using the same spatial coordinates to identify plot change information; Based on the land use change information, determine the farmland use change pattern, calculate the occurrence area and spatial distribution of each farmland use change pattern, and obtain classified change data; Based on the classified change data, the change time node of each changed plot is calculated, the monthly interval of the change is determined by the acquisition time of the multi-temporal optical remote sensing image, and farmland use change data is established by combining the boundary coordinates and area information of the changed plots. Based on the farmland use change data, key areas of concern with abnormal change frequencies are identified, and change statistics, trend analysis results, and early warning area information are summarized to generate a farmland grain planting status monitoring report.

[0012] Secondly, the present invention provides a dynamic monitoring system for grain cultivation on arable land, the dynamic monitoring system for grain cultivation on arable land comprising: The acquisition module is used to acquire multi-temporal optical remote sensing images of the target area and reconstruct the multi-temporal optical remote sensing images to obtain high-resolution farmland images. The crop identification module is used to extract features and identify crops from the high-resolution farmland image to obtain a crop type distribution map; The generation module is used to perform multi-time comparative analysis on the crop type distribution map to obtain farmland use change data, and generate a farmland grain planting monitoring report based on the farmland use change data.

[0013] A third aspect of the present invention provides a computer-readable storage medium storing instructions that, when executed on a computer, cause the computer to perform the above-described method for dynamic monitoring of farmland grain planting conditions.

[0014] The technical solution provided by this invention utilizes farmland block sparse channel estimation technology to fully leverage the regular geometric distribution characteristics of farmland blocks to construct a dedicated sparse base dictionary. Combined with crop phenological weight factors, the sparse reconstruction parameters are dynamically adjusted, effectively solving the quality degradation problem of satellite remote sensing images during transmission. A dual-drive super-resolution reconstruction mechanism, employing boundary preservation and spectral feature enhancement, is used to improve image resolution while maintaining the clarity of farmland block boundaries, with a focus on enhancing the spectral distinguishing features between food crops and cash crops. An adaptive farmland scale mechanism dynamically adjusts the convolution kernel size based on the block area, effectively addressing the technical challenge of traditional methods being unable to adapt to farmland blocks of different scales. A temporal attention mechanism is used to fuse multi-temporal crop growth features, combined with prior crop phenological knowledge to guide temporal weight allocation, fully exploring the temporal patterns of crop growth. Compared to static image analysis methods, this invention achieves higher crop identification accuracy. This invention employs spectral feature separation and crop feature enhancement technologies to achieve accurate differentiation between food crops such as rice, wheat, and corn and cash crops, effectively identifying subtle "non-food" changes and overcoming the limitation of traditional methods that can only detect obvious land use transitions. Attached Figure Description

[0015] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a schematic diagram illustrating the steps of the dynamic monitoring method for grain cultivation on cultivated land in an embodiment of the present invention; Figure 2 This is a schematic diagram of the structure of the dynamic monitoring system for grain cultivation on cultivated land in an embodiment of the present invention. Detailed Implementation

[0017] This invention provides a method, system, and storage medium for dynamic monitoring of grain cultivation in arable land. The terms "first," "second," "third," "fourth," etc. (if present)," in the specification, claims, and accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in a sequence other than that illustrated or described herein. Furthermore, the terms "comprising" or "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0018] For ease of understanding, the specific process of the embodiments of the present invention is described below. Please refer to [link / reference]. Figure 1 One embodiment of the dynamic monitoring method for grain cultivation on cultivated land in this invention includes: Step S1: Acquire multi-temporal optical remote sensing images of the target area and reconstruct the multi-temporal optical remote sensing images to obtain high-resolution farmland images. In this embodiment, multi-temporal optical remote sensing images are acquired during key phenological stages in the target area, including sowing, seedling, jointing, heading, grain-filling, and maturity stages. After geometric, radiometric, and atmospheric corrections, all temporal data are registered in a unified coordinate system. Spatial feature analysis of farmland plots is performed on the remote sensing images, extracting the geometric boundary shapes, spatial arrangement patterns, and structural features of the plots. These spatial basis functions are combined to construct a sparse basis dictionary for farmland, effectively representing prior knowledge of the plotted distribution of farmland. Simultaneously, the spectral texture response patterns of different crops in multiple bands (red, green, blue, red-edge, near-infrared, and shortwave infrared) in the multi-temporal images are statistically analyzed. A farmland texture correlation matrix is ​​constructed using spatiotemporal variation characteristics, revealing the spectral differences and spatial relationships of different crops at different stages. The crop growth period is identified by combining the image acquisition time, and phenological weight factors are calculated using the physiological changes of crops at each key stage. This allows the reconstruction algorithm to adapt to the characteristics of crops at different stages, and the sparsity threshold is dynamically determined through the phenological weight factors. Based on a sparse base dictionary of farmland, a farmland texture correlation matrix, and a sparsity threshold, degraded channels during image transmission are iteratively reconstructed to progressively optimize the channel response, resulting in a channel response matrix that characterizes the true spectral and spatial properties of crops. This channel response matrix is ​​then input into a dual-drive super-resolution reconstruction network. Farmland boundary preservation is used to ensure the continuity and clarity of plot boundaries, while non-grain crop detection is used to enhance the distinguishing features between grain and non-grain crops, generating high-resolution farmland images.

[0019] Step S2: Extract features and identify crops from high-resolution farmland images to obtain a crop type distribution map; In this embodiment, area statistical analysis is performed on each farmland plot in the high-resolution farmland image. Scale-matching features are constructed using parameters such as the mean, standard deviation, and quantiles of the statistical area distribution. These scale-matching features are input into a multi-scale feature pyramid structure, and multi-scale spatial features containing detailed information and global semantics are extracted through parallel processing of convolutional kernels at different scales. This approach takes into account both the fine texture of small plots and the overall spatial pattern of large plots. Combining the temporal series information of multi-temporal optical remote sensing images, the multi-scale spatial features are input into a temporal attention module. A self-attention mechanism is used to weightedly fuse features from different temporal phases, generating temporal enhancement features that reflect the dynamic changes in crop growth cycles. This enables the model to identify changes in spectral reflectance and spatial texture of crops at different phenological stages. Based on the temporal enhancement features, and combined with the geometric boundary information of farmland plots and the phenological characteristics of crops at different growth stages, the feature representation is corrected to maintain clear farmland boundaries and make crop category features more representative, thus forming a multi-scale feature map of the farmland. A deep learning-based classification and segmentation model is used to identify crops in multi-scale feature maps of farmland, distinguishing major food crops such as rice, wheat, and corn from cash crops, forest land, or abandoned plots. The model also generates crop type distribution maps at the plot level, reflecting the crop planting pattern of the target area.

[0020] Step S3: Perform multi-time comparative analysis on the crop type distribution map to obtain farmland use change data, and generate a farmland grain planting monitoring report based on the farmland use change data.

[0021] In this embodiment, crop type distribution maps from different time periods are compared pixel-by-pixel under a unified spatial coordinate system. Pixel-level difference detection identifies plot changes, determining whether a plot has undergone a transformation from food crops to cash crops, forest land, construction land, or abandonment during different monitoring periods. Based on plot change information, arable land use change patterns are categorized and identified, establishing a set of change patterns including food-cash crop conversion, arable land abandonment, agricultural-to-non-agricultural conversion, and agricultural-to-forest conversion. The area and spatial distribution of each change pattern are calculated, forming classified change data with spatiotemporal distribution characteristics. Based on the classified change data and combined with the acquisition time of multi-temporal optical remote sensing images, the change time node is calculated for each plot that has changed. Time series analysis methods are used to determine the monthly interval of the change. Simultaneously, the plot's boundary coordinates, area information, and change type are stored together, forming structured arable land use change data. By statistically analyzing the frequency of changes in cultivated land use within a region using data on changes in cultivated land use, key areas of concern with abnormal frequency of changes are identified. Through aggregated calculations, statistical results and trend analysis data on changes are obtained, and areas at risk of "non-grain" or "non-agricultural" conversion are marked with early warnings. The statistical data on changes, trend analysis results, and information on early warning areas are then compiled and organized to generate a monitoring report on cultivated land use for grain production.

[0022] In one specific embodiment, the process of performing step S1 may specifically include the following steps: Acquire multi-temporal optical remote sensing images of the target area; The spatial basis functions of farmland plots in multi-temporal optical remote sensing images are analyzed, and the spatial basis functions are combined to form a sparse basis dictionary for farmland. The texture response patterns of different crop types in various spectral bands in multi-temporal optical remote sensing images were statistically analyzed, and a farmland texture correlation matrix was constructed based on the texture response patterns. Acquire temporal information from multi-temporal optical remote sensing images and identify the corresponding crop growth stages. Calculate crop phenological weight factors based on crop growth stages and determine sparsity thresholds using crop phenological weight factors. Based on the sparse base dictionary of farmland, the correlation matrix of farmland texture, and the sparsity threshold, the transmission channel of multi-temporal optical remote sensing images is iteratively reconstructed to obtain the channel response matrix. High-resolution farmland images are obtained by performing dual-drive super-resolution reconstruction on multi-temporal optical remote sensing images based on the channel response matrix.

[0023] In this embodiment, multi-temporal optical remote sensing images of the target area are acquired during key growth stages, including sowing, seedling, jointing, heading, grain-filling, and maturity. Image types include high-resolution satellite imagery, multispectral imagery, and hyperspectral imagery. Geometric, radiometric, and atmospheric corrections are performed to eliminate interference from sensor noise, illumination variations, and atmospheric scattering, ensuring that images from different time phases and sources can be registered in a unified coordinate system. The spatial geometric features of farmland plots in the remote sensing images are analyzed, extracting the geometric shape, boundary orientation, and regularized distribution characteristics of each plot. A set of spatial basis functions is generated using methods such as ellipse fitting and rectangularization. These spatial basis functions are combined to form a sparse basis dictionary for farmland, serving as a priori knowledge base for sparse image reconstruction, thereby ensuring the authenticity of farmland boundaries and spatial layout during image reconstruction. Statistical analysis and modeling of texture response patterns of different crop types across multiple spectral bands in imagery are performed. For example, differences in chlorophyll content are analyzed in the red-edge band, crop biomass characteristics are analyzed in the near-infrared band, and changes in crop water content are identified in the short-wave infrared band. These cross-band texture response patterns are then integrated to construct a farmland texture correlation matrix, reflecting the spectral consistency and differences of crop categories during spatiotemporal evolution, providing spectral dimensional constraints for sparse channel estimation. The corresponding crop growth stages are identified by combining the acquisition time information of remote sensing images, and crop phenological weight factors are calculated based on the physiological characteristics changes at each growth stage. These phenological weight factors are modeled using an exponential decay function, resulting in higher weights during key stages such as heading and maturity, and relatively lower weights during stages such as seedling and grain-filling. This weight adjustment mechanism dynamically determines the sparsity threshold. By utilizing a sparse basis dictionary for farmland, a farmland texture correlation matrix, and a dynamically adjusted sparsity threshold, the transmission channel of multi-temporal remote sensing images is iteratively reconstructed. An improved orthogonal matching pursuit algorithm is used to progressively optimize channel estimation. During the iteration process, basis functions that most closely resemble the geometry of real farmland and the spectral curves of crops are continuously selected. The process stops when the residual energy decreases to a preset threshold or the maximum number of iterations is reached, resulting in a channel response matrix that characterizes the spatial structure of real farmland and the spectral features of crops. Based on the channel response matrix, dual-drive super-resolution reconstruction of the remote sensing images is performed. Boundary preservation drive maintains the clarity and continuity of farmland plot boundaries through a boundary preservation loss function, while non-grain detection drive enhances the spectral and spatial distinguishability between food crops, cash crops, forest land, and non-agricultural areas through a crop segmentation loss function. The network employs progressive upsampling combined with a multi-scale boundary refinement module to improve resolution while maintaining the integrity of farmland texture features. Boundary representation and crop recognition capabilities are progressively optimized during multi-stage training, outputting high-resolution farmland images.

[0024] In one specific embodiment, the process of performing dual-drive super-resolution reconstruction of multi-temporal optical remote sensing images based on the channel response matrix to obtain high-resolution farmland images can specifically include the following steps: A sparse reconstruction model is established using a sparse basis dictionary of farmland and a farmland texture correlation matrix, and the transmission channel of multi-temporal optical remote sensing images is represented as the initial sparse processing result of sparse basis functions and sparse coefficients. Candidate basis functions are selected one by one from the spatial basis functions of the sparse basis dictionary of farmland. The similarity between the candidate basis functions and the known farmland plot shapes is calculated. Basis functions with similarity greater than a preset threshold are selected as structural constraint basis functions. The initial sparse processing result is reconstructed based on the sparse reconstruction model and structural constraint basis functions to generate the reconstruction result; Based on the reconstruction results and the characteristic spectral curves, the candidate basis function with the smallest spectral angular distance is selected as the optimal basis function, and the residual values ​​are updated. The number of iterations is controlled by the sparsity threshold. The method is to repeatedly select the method that minimizes the residual value and satisfies the optimal basis function until the residual converges or reaches the sparsity threshold, thus obtaining the channel response matrix.

[0025] In this embodiment, a sparse reconstruction model is established based on a sparse basis dictionary of farmland and a farmland texture correlation matrix. The degradation manifestation of multi-temporal optical remote sensing images in the transmission channel is represented by basis function expansion and sparse coefficients as the initial sparse processing result. The initial sparse processing result is equivalent to projecting the complex signal degradation process onto the space of the sparse basis dictionary, so that the main structure and texture pattern of the original channel can be approximated by a finite number of basis functions. Candidate basis functions are selected one by one from the spatial basis functions of the sparse basis dictionary, and the similarity between each candidate basis function and the known farmland plot geometry is calculated. The similarity is measured by a combination of methods such as shape context descriptors and Hausdorff distance. When the candidate basis function and the plot boundary have a high degree of consistency in spatial topology and geometric distribution, the similarity value is greater than a preset threshold. At this time, the candidate basis function is screened as a structural constraint basis function. Only basis functions that meet the spatial geometric consistency constraint will participate in the subsequent reconstruction calculation, avoiding the accumulation of reconstruction errors due to the introduction of irrelevant basis functions. Within the framework of the sparse reconstruction model, the initial sparse processing results are progressively reconstructed using pre-selected structural constraint basis functions, generating the reconstruction result for the current iteration. The reconstruction result of each iteration is compared with the characteristic spectral curves of different crop categories, using spectral angular distance as the discrimination index. The spectral similarity is measured by calculating the angle between the spectral features of the reconstructed result and the standard crop spectral curve. The candidate basis function with the smallest spectral angular distance is selected as the optimal basis function from the candidate basis function set, and the current residual value is updated using the optimal basis function. The residual value reflects the magnitude of the difference between the original observed signal and the current reconstructed signal. As optimal basis functions are continuously selected and superimposed, the residual value gradually decreases. To control the iteration process, a sparsity threshold is introduced as a stopping condition. In each iteration, the algorithm selects the optimal basis function that minimizes the residual value and satisfies both geometric and spectral constraints, continuously updating the reconstruction result and residual until the residual value converges below the preset convergence threshold, or until the maximum number of iterations set by the sparsity threshold is reached. After iteration, the channel response matrix is ​​obtained.

[0026] In one specific embodiment, the process of performing dual-drive super-resolution reconstruction of multi-temporal optical remote sensing images based on the channel response matrix to obtain high-resolution farmland images can specifically include the following steps: The channel response matrix and multi-temporal optical remote sensing images are input into the dual-drive super-resolution reconstruction network. Multi-scale feature extraction is performed through the encoder in the dual-drive super-resolution reconstruction network to obtain the encoded feature map. Boundary preservation is performed based on the encoded feature map to generate a boundary enhancement feature map, and spectral feature enhancement is performed on the boundary enhancement feature map to obtain a spectral enhancement feature map; The spectral enhancement feature map is upsampled and reconstructed by the decoder in the dual-drive super-resolution reconstruction network, while maintaining the plot boundary clarity of the boundary enhancement feature map and the crop differentiation features of the spectral enhancement feature map, to obtain high-resolution farmland imagery.

[0027] In this embodiment, the channel response matrix and multi-temporal optical remote sensing images are jointly input into a dual-drive super-resolution reconstruction network. Feature extraction is performed in the front-end encoder section of the network using a multi-scale convolutional structure. The encoder employs a combination of residual dense blocks and channel attention mechanisms, enabling lower-level convolutions to capture fine-grained texture features, while higher-level convolutions obtain global semantic representations, forming an encoded feature map that preserves the spatial distribution information of farmland plots and the spectral response information of crops. Based on the encoded feature map, a boundary-preserving drive is executed. The boundary-preserving drive module extracts farmland plot boundaries using an improved edge detection operator and enhances edge features by combining gradient loss and total variational regularization, generating an enhanced boundary feature map that amplifies the weight of plot boundaries in the feature space. Following boundary enhancement, a spectral feature enhancement stage is introduced. The spectral enhancement module utilizes the spectral characteristics of crops in key bands such as red light, red edge, near-infrared, and short-wave infrared to perform channel-by-channel weighted correction of the feature space, making the differences between food crops and cash crops more pronounced in the spectral dimension, generating an enhanced spectral feature map. The spectral enhancement feature map is input into the decoder for upsampling reconstruction. The decoder uses a progressive upsampling method combined with sub-pixel convolution and bilinear interpolation to gradually restore the spatial resolution. Residual connections of boundary enhancement features are introduced in each upsampling stage to continuously enhance the boundary sharpness during the resolution restoration process, resulting in high-resolution farmland images.

[0028] In one specific embodiment, the process of inputting the channel response matrix and multi-temporal optical remote sensing images into a dual-drive super-resolution reconstruction network, and extracting multi-scale features through the encoder in the dual-drive super-resolution reconstruction network to obtain the encoded feature map can specifically include the following steps: Channel-corrected images are obtained by using the channel response matrix to compensate for transmission distortion in multi-temporal optical remote sensing images. The channel-corrected image is input into the feature extraction module of the encoder in the dual-drive super-resolution reconstruction network to extract features and obtain the original features at each scale. Channel attention processing is performed on the original features at each scale to obtain channel-enhanced features; Based on the quality assessment information of the channel response matrix, the channel enhancement features are adaptively adjusted, and the encoded feature map is output.

[0029] In this embodiment, the channel response matrix is ​​used to compensate for transmission distortion in multi-temporal optical remote sensing images. The channel response matrix reflects the true transmission characteristics and degradation patterns of remote sensing signals in different bands. By performing pixel-by-pixel convolution or deconvolution operations between the channel response matrix and the image data, compensation is achieved for blurring, noise, and offset caused by transmission distortion in the image, resulting in a channel-corrected image. The channel-corrected image is input into the encoder part of the dual-drive super-resolution reconstruction network. The encoder's feature extraction module consists of multi-scale convolutional layers and residual dense blocks, which can simultaneously capture low-level texture details and high-level spatial structure information in the image. The original features at each scale are extracted through hierarchical convolution. Channel attention processing is performed on the original features at each scale. The channel attention mechanism obtains the importance distribution of different channels by performing global average pooling and weighted calculation on each feature channel, and generates weight coefficients using fully connected layers and activation functions to highlight channels related to crop identification and boundary preservation in the feature map, while suppressing background noise or interference from non-agricultural land. Simultaneously, considering the quality differences within the channel response matrix itself, the channel enhancement features are adaptively adjusted based on the quality assessment information of the channel response matrix. This quality assessment information includes indicators such as mean square error, correlation coefficient, and spectral similarity, which are used to determine the reliability of image correction for a specific portion. If the channel quality assessment is excellent, the channel enhancement features retain their original weights; if the assessment is good, further smoothing is applied to the channel attention output to reduce noise sensitivity; if the assessment is moderate or poor, regularization constraints are added to the feature channels, or redundant corrections are introduced in the decoding stage. Through this adaptive adjustment process, the encoder can output boundary-sensitive geometric features and spectral features effective for crop classification, and can automatically optimize feature weights based on channel quality differences, avoiding a decrease in overall recognition accuracy due to local channel distortion.

[0030] In one specific embodiment, the process of performing step S2 may specifically include the following steps: Scale-matching features were obtained by statistically analyzing the area of ​​farmland plots in high-resolution farmland images. The scale-matching features are input into a multi-scale feature pyramid for parallel processing to obtain multi-scale spatial features. Based on the time series information of multi-temporal optical remote sensing images, multi-scale spatial features are input into the temporal attention module for temporal dimension fusion to obtain temporal enhanced features; Based on the temporal enhancement features, feature correction is performed by combining the geometric boundary information of farmland plots and crop phenological characteristics to obtain a multi-scale feature map of farmland; Crop identification is performed based on multi-scale feature maps of farmland to obtain crop type distribution maps.

[0031] In this embodiment, area statistical analysis is performed on farmland plots in high-resolution farmland images. By calculating the number of pixels and actual area of ​​each plot, statistical quantities such as the mean, variance, and quantiles of the overall plot area are obtained, and scale-matching features are generated from these. Scale-matching features reflect the spatial distribution patterns and area differences of farmland in the target area, enabling the feature extraction process to adopt appropriate convolutional kernel sizes and feature fusion strategies according to plots of different sizes, thereby avoiding the problem of lost details in small plots or insufficient features in large plots. The scale-matching features are input into a multi-scale feature pyramid structure for parallel processing. The multi-scale feature pyramid structure sets multiple scale layers, each using convolutional kernels of different sizes and feature channels to extract image features. This allows the network to simultaneously capture local texture details and global spatial semantic information of plots at the same level, resulting in multi-scale spatial features. By combining the temporal series information of multi-temporal optical remote sensing images, multi-scale spatial features are input into a temporal attention module. A self-attention mechanism is used to weight features from different time phases, emphasizing stages with significant temporal differences, such as sowing and maturity, while reducing the weight of stages with smaller spectral differences, such as seedling and grain-filling stages. This enhances the characteristics of the crop growth cycle in the temporal dimension, resulting in temporally enhanced features. Based on these enhanced features, feature correction is performed by combining the geometric boundary information of farmland plots and crop phenological characteristics. Geometric boundary information ensures the clarity and continuity of farmland boundaries during feature representation, avoiding boundary blurring caused by upsampling or convolution operations. Crop phenological characteristics are dynamically weighted based on the spectral responses of different crops at various growth stages, making the feature representation more consistent with the actual physiological changes of crops, thus generating a multi-scale feature map of farmland. Crop identification is performed based on multi-scale feature maps of farmland. A deep learning classification and segmentation model is used to distinguish the crop category of each plot, differentiating major food crops such as rice, wheat, and corn from cash crops, forest land, or abandoned plots, and generating a crop type distribution map at the plot level.

[0032] In one specific embodiment, the process of performing crop identification based on a multi-scale feature map of farmland to obtain a crop type distribution map may specifically include the following steps: The multi-scale feature map of farmland is input into the spectral feature separation layer of the crop recognition model to separate spectral components and obtain the spectral separation result; Crop feature enhancement is performed based on the grain crop components in the spectral separation results to obtain enhanced spectral features; The enhanced spectral features are input into the crop classification layer of the crop identification model for type identification, and the crop classification results are obtained. Based on the crop classification results, a connectivity analysis is performed, and a crop type distribution map is generated by combining farmland plot boundary information.

[0033] In this embodiment, a multi-scale feature map of farmland is input into the spectral feature separation layer of the crop recognition model. The spectral feature separation layer uses algorithms such as independent component analysis or non-negative matrix factorization to decompose the spectral information in the input features according to the band response patterns of different crops, obtaining spectral separation results. During this process, the complex mixed spectral signal is divided into subspaces such as food crop components, cash crop components, and background components, ensuring that the features of different crops in the spectral dimension can be expressed separately, avoiding category uncertainty caused by mixed pixels. Food crop components are extracted from the spectral separation results, and crop feature enhancement is performed on these components. The enhancement module weights and amplifies the key spectral responses of food crops in the red-edge band, near-infrared band, and short-wave infrared band, making the performance of major food crops such as rice, wheat, and corn more prominent in the feature space, while suppressing the interference of cash crops and non-agricultural areas in these key bands, resulting in enhanced spectral features. The enhanced spectral features are input into the crop classification layer of the crop recognition model. The classification layer is composed of a convolutional neural network or a deep residual network, extracting local patterns through convolution operations and calculating category probabilities through fully connected layers or a softmax function, outputting the crop classification results. The crop identification model can provide a predicted label for each pixel or plot unit, assigning it to a specific crop category. It fully utilizes the improved discriminative power of spectral enhancement during category discrimination, thereby increasing the classification accuracy between food and non-food crops. Connectivity analysis is performed based on the crop classification results to identify spatial aggregation features and avoid misclassification of scattered small regions due to local noise. Through connectivity analysis, spatially adjacent pixels belonging to the same category are merged into a unified region, and spatial correction is performed using farmland plot boundary information. This maintains the integrity of plot boundaries while correcting misclassified pixels near the boundaries, generating a crop type distribution map.

[0034] In one specific embodiment, the process of performing step S3 may specifically include the following steps: By comparing crop type distribution maps from different time periods pixel by pixel using the same spatial coordinates, information on plot changes can be identified. Based on the information on changes in land parcels, determine the patterns of farmland use change, and calculate the area and spatial distribution of each farmland use change pattern to obtain classified change data; The change time points for each changed plot are calculated based on the classification change data. The monthly intervals in which the changes occurred are determined by the acquisition time of multi-temporal optical remote sensing images. Farmland use change data are established by combining the boundary coordinates and area information of the changed plots. Based on the data on changes in cultivated land use, key areas of concern with abnormal frequency of change are identified, and statistical data on changes, trend analysis results, and information on early warning areas are summarized to generate a monitoring report on the situation of grain cultivation on cultivated land.

[0035] In this embodiment, crop type distribution maps from different time phases are registered under a unified spatial coordinate system. Difference detection is performed on the category label of each pixel. If a pixel changes its category from food crops to cash crops, forest land, construction land, or abandoned land in two or more time phases, it is determined that the corresponding pixel has undergone land use change, thus identifying land use change information. Based on the land use change information, the change pattern of cultivated land use is determined. Different change patterns include food-cash crop conversion, agricultural-to-non-agricultural conversion, agricultural-to-forest conversion, and farmland abandonment. By aggregating and statistically analyzing pixels of various change patterns, the occurrence area and spatial distribution of different patterns within the region are calculated, obtaining categorized change data with classification and spatiotemporal attributes, reflecting the dynamic pattern of cultivated land use at different time stages. Based on the categorized change data, the change time node corresponding to each changed plot is calculated. By comparing the category differences across different time phases, the start and end times of the change are determined. Using the acquisition time of multi-temporal optical remote sensing images as a time reference, the change event is located to a specific monthly interval. By combining the boundary coordinates and area information of the changed plots, a farmland use change data archive is established for each plot. The archive includes elements such as plot boundary shape, area size, change category, and time interval. Based on the farmland use change data, time-series statistics and frequency analysis are performed on the entire region to identify key areas of concern that have experienced multiple changes or abnormally high change rates within a short period. All farmland use change data are summarized to generate a comprehensive output that includes change statistics, trend analysis results, and early warning area information, forming a farmland grain cultivation monitoring report.

[0036] The above describes the method for dynamic monitoring of farmland grain cultivation in embodiments of the present invention. The following describes the system for dynamic monitoring of farmland grain cultivation in embodiments of the present invention. Please refer to [link / reference]. Figure 2 One embodiment of the dynamic monitoring system for grain cultivation on cultivated land in this invention includes: The acquisition module is used to acquire multi-temporal optical remote sensing images of the target area and reconstruct the multi-temporal optical remote sensing images to obtain high-resolution farmland images. The crop identification module is used for feature extraction and crop identification of high-resolution farmland images to obtain crop type distribution maps; The generation module is used to perform multi-time comparative analysis on crop type distribution maps, obtain farmland use change data, and generate a farmland grain planting monitoring report based on the farmland use change data.

[0037] Through the collaborative efforts of the aforementioned components, this invention utilizes farmland block sparse channel estimation technology to fully leverage the regular geometric distribution characteristics of farmland blocks to construct a dedicated sparse base dictionary. Combined with crop phenological weight factors, the sparse reconstruction parameters are dynamically adjusted, effectively solving the quality degradation problem of satellite remote sensing images during transmission. A dual-drive super-resolution reconstruction mechanism, employing boundary preservation and spectral feature enhancement, improves image resolution while maintaining the clarity of farmland block boundaries, and focuses on enhancing the spectral distinguishing features between food crops and cash crops. An adaptive farmland scale mechanism dynamically adjusts the convolution kernel size based on the block area, effectively addressing the technical challenge of traditional methods being unable to adapt to farmland blocks of different scales. A temporal attention mechanism is used to fuse multi-temporal crop growth features, combined with prior crop phenological knowledge to guide temporal weight allocation, fully exploring the temporal patterns of crop growth. Compared to static image analysis methods, this invention achieves higher crop identification accuracy. This invention employs spectral feature separation and crop feature enhancement technologies to achieve accurate differentiation between food crops such as rice, wheat, and corn and cash crops, effectively identifying subtle "non-food" changes and overcoming the limitation of traditional methods that can only detect obvious land use transitions.

[0038] An embodiment of the present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-described method. It is understood that the computer-readable storage medium in this embodiment can be a volatile readable storage medium or a non-volatile readable storage medium.

[0039] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the present invention and embodiments can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual-rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM, etc.

[0040] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0041] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0042] The above-described 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 the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for dynamic monitoring of grain cultivation on arable land, characterized in that, include: Acquiring multi-temporal optical remote sensing images of a target area and reconstructing them to obtain high-resolution farmland images; specifically including: acquiring multi-temporal optical remote sensing images of the target area; analyzing the spatial basis functions of farmland plots in the multi-temporal optical remote sensing images and combining the spatial basis functions to form a sparse basis dictionary for farmland; statistically analyzing the texture response patterns of different crop types in each spectral band in the multi-temporal optical remote sensing images and constructing a farmland texture correlation matrix based on the texture response patterns; acquiring the temporal information of the multi-temporal optical remote sensing images and identifying the corresponding crop growth stages, calculating crop phenological weight factors based on the crop growth stages, and determining a sparsity threshold using the crop phenological weight factors; iteratively reconstructing the transmission channels of the multi-temporal optical remote sensing images based on the farmland sparse basis dictionary, the farmland texture correlation matrix, and the sparsity threshold to obtain a channel response matrix; and performing dual-drive super-resolution reconstruction of the multi-temporal optical remote sensing images based on the channel response matrix to obtain high-resolution farmland images. Feature extraction and crop identification are performed on the high-resolution farmland images to obtain a crop type distribution map; A multi-time comparative analysis was performed on the crop type distribution map to obtain farmland use change data, and a farmland grain planting monitoring report was generated based on the farmland use change data.

2. The method for dynamic monitoring of grain cultivation on cultivated land according to claim 1, characterized in that, The step of performing dual-drive super-resolution reconstruction on the multi-temporal optical remote sensing image based on the channel response matrix to obtain a high-resolution farmland image includes: A sparse reconstruction model is established using the farmland sparse basis dictionary and the farmland texture correlation matrix, and the transmission channel of the multi-temporal optical remote sensing image is represented as the initial sparse processing result of sparse basis functions and sparse coefficients. Candidate basis functions are selected one by one from the spatial basis functions of the farmland sparse basis dictionary. The similarity between the candidate basis functions and the known farmland plot shapes is calculated. Basis functions with similarity greater than a preset threshold are selected as structural constraint basis functions. Based on the sparse reconstruction model and the structural constraint basis function, the initial sparse processing result is reconstructed to generate the reconstruction result; Based on the reconstruction results and the characteristic spectral curves, the candidate basis function with the smallest spectral angular distance is selected as the optimal basis function, and the residual values ​​are updated. The number of iterations is controlled according to the sparsity threshold. The method is to repeatedly select the method that minimizes the residual value and satisfies the optimal basis function until the residual converges or reaches the sparsity threshold, thereby obtaining the channel response matrix.

3. The method for dynamic monitoring of grain cultivation on cultivated land according to claim 2, characterized in that, The step of performing dual-drive super-resolution reconstruction on the multi-temporal optical remote sensing image based on the channel response matrix to obtain a high-resolution farmland image includes: The channel response matrix and the multi-temporal optical remote sensing image are input into a dual-drive super-resolution reconstruction network. Multi-scale feature extraction is performed through the encoder in the dual-drive super-resolution reconstruction network to obtain a coded feature map. Based on the encoded feature map, a boundary preservation driving process is performed to generate a boundary enhancement feature map, and the boundary enhancement feature map is then subjected to spectral feature enhancement to obtain a spectral enhancement feature map. The spectral enhancement feature map is upsampled and reconstructed by the decoder in the dual-drive super-resolution reconstruction network, while maintaining the plot boundary clarity of the boundary enhancement feature map and the crop differentiation features of the spectral enhancement feature map, to obtain a high-resolution farmland image.

4. The method for dynamic monitoring of grain cultivation on cultivated land according to claim 3, characterized in that, The process involves inputting the channel response matrix and the multi-temporal optical remote sensing image into a dual-drive super-resolution reconstruction network, and then performing multi-scale feature extraction through the encoder in the dual-drive super-resolution reconstruction network to obtain an encoded feature map, including: The channel response matrix is ​​used to compensate for transmission distortion in the multi-temporal optical remote sensing image to obtain a channel-corrected image. The channel-corrected image is input into the feature extraction module of the encoder in the dual-drive super-resolution reconstruction network for feature extraction to obtain the original features at each scale. Channel attention processing is performed on the original features at each scale to obtain channel-enhanced features; The channel enhancement features are adaptively adjusted based on the quality assessment information of the channel response matrix, and the encoded feature map is output.

5. The method for dynamic monitoring of grain cultivation on cultivated land according to claim 1, characterized in that, The process of extracting features and identifying crops from the high-resolution farmland images to obtain a crop type distribution map includes: Scale matching features were obtained by performing area statistical analysis on farmland plots in the high-resolution farmland image. The scale-matching features are input into a multi-scale feature pyramid for parallel processing to obtain multi-scale spatial features. Based on the time series information of the multi-temporal optical remote sensing images, the multi-scale spatial features are input into the temporal attention module for temporal dimension fusion to obtain temporal enhancement features; Based on the aforementioned temporal enhancement features, feature correction is performed by combining the geometric boundary information of farmland plots and crop phenological characteristics to obtain a multi-scale feature map of farmland. Crop identification is performed based on the multi-scale feature map of the farmland to obtain a crop type distribution map.

6. The method for dynamic monitoring of grain cultivation on cultivated land according to claim 5, characterized in that, The crop identification based on the multi-scale feature map of the farmland to obtain a crop type distribution map includes: The multi-scale feature map of the farmland is input into the spectral feature separation layer of the crop recognition model to separate spectral components and obtain the spectral separation result. Based on the grain crop components in the spectral separation results, crop feature enhancement is performed to obtain enhanced spectral features; The enhanced spectral features are input into the crop classification layer of the crop recognition model for type identification to obtain crop classification results. Based on the crop classification results, a connected region analysis is performed, and a crop type distribution map is generated by combining farmland plot boundary information.

7. The method for dynamic monitoring of grain cultivation on cultivated land according to claim 1, characterized in that, The process involves performing a multi-time comparative analysis of the crop type distribution map to obtain farmland use change data, and generating a farmland grain planting monitoring report based on the farmland use change data, including: The distribution maps of the crop types at different time points are compared pixel by pixel using the same spatial coordinates to identify plot change information; Based on the land use change information, determine the farmland use change pattern, calculate the occurrence area and spatial distribution of each farmland use change pattern, and obtain classified change data; Based on the classified change data, the change time node of each changed plot is calculated, the monthly interval of the change is determined by the acquisition time of the multi-temporal optical remote sensing image, and farmland use change data is established by combining the boundary coordinates and area information of the changed plots. Based on the farmland use change data, key areas of concern with abnormal change frequencies are identified, and change statistics, trend analysis results, and early warning area information are summarized to generate a farmland grain planting status monitoring report.

8. A dynamic monitoring system for grain cultivation on arable land, characterized in that, The method for dynamically monitoring the grain planting situation on cultivated land as described in any one of claims 1-7 includes: An acquisition module is used to acquire multi-temporal optical remote sensing images of a target area and reconstruct the multi-temporal optical remote sensing images to obtain high-resolution farmland images. Specifically, this includes: acquiring multi-temporal optical remote sensing images of the target area; analyzing the spatial basis functions of farmland plots in the multi-temporal optical remote sensing images and combining the spatial basis functions to form a sparse basis dictionary for farmland; statistically analyzing the texture response patterns of different crop types in each spectral band of the multi-temporal optical remote sensing images and constructing a farmland texture correlation matrix based on the texture response patterns; acquiring the temporal information of the multi-temporal optical remote sensing images and identifying the corresponding crop growth stages, calculating crop phenological weight factors based on the crop growth stages, and determining a sparsity threshold using the crop phenological weight factors; iteratively reconstructing the transmission channel of the multi-temporal optical remote sensing images based on the farmland sparse basis dictionary, the farmland texture correlation matrix, and the sparsity threshold to obtain a channel response matrix; and performing dual-drive super-resolution reconstruction of the multi-temporal optical remote sensing images based on the channel response matrix to obtain high-resolution farmland images. The crop identification module is used to extract features and identify crops from the high-resolution farmland image to obtain a crop type distribution map; The generation module is used to perform multi-time comparative analysis on the crop type distribution map to obtain farmland use change data, and generate a farmland grain planting monitoring report based on the farmland use change data.

9. A computer-readable storage medium, characterized in that, It stores a computer program, which, when run by a processor, causes the processor to execute the dynamic monitoring method for grain cultivation on cultivated land as described in any one of claims 1 to 7.

Images