Method and system for monitoring the situation of ploughing and planting, and storage medium

Abstract

The application relates to the technical field of intelligent monitoring, and discloses a farmland planting situation dynamic monitoring method and system and a storage medium, the method comprising the following steps: acquiring multi-temporal optical remote sensing images of a target area, and reconstructing the multi-temporal optical remote sensing images to obtain high-resolution farmland images; performing feature extraction and crop identification on the high-resolution farmland images to obtain a crop type distribution map; performing multi-temporal comparison analysis on the crop type distribution map to obtain farmland utilization change data, and generating a farmland planting situation monitoring report based on the farmland utilization change data; the application effectively solves the quality degradation problem of satellite remote sensing images in the transmission process, realizes accurate differentiation of rice, wheat, corn and other food crops and economic crops, effectively identifies subtle non-grain changes, and improves the accuracy of farmland planting situation dynamic monitoring.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent monitoring, and in particular to a farmland planting situation dynamic monitoring method and system and a storage medium. BACKGROUND

[0002] Traditional farmland monitoring methods rely on manual field investigation and simple remote sensing image analysis, which has problems such as long monitoring period, limited coverage, high labor cost, and is difficult to meet the requirements of large-scale, high-frequency dynamic monitoring. At the same time, the existing remote sensing monitoring technology does not fully consider the particularity of farmland scenes when dealing with complex agricultural landscapes, lacks targeted solutions to the quality degradation problem in the satellite signal transmission process, and ignores the special requirements of crop identification accuracy for agricultural monitoring tasks, resulting in that although the reconstructed image is visually improved, there are still deficiencies in crop type differentiation and boundary identification, which affects the detection accuracy of crop identification and farmland change. SUMMARY

[0003] The present application provides a farmland planting situation dynamic monitoring method and system and a storage medium, which effectively solves the quality degradation problem of satellite remote sensing images in the transmission process, realizes accurate differentiation of food crops such as rice, wheat, and corn and economic crops, effectively identifies subtle non-food changes, and improves the accuracy of farmland planting situation dynamic monitoring.

[0004] In a first aspect, the present application provides a farmland planting situation dynamic monitoring method, which comprises:

[0005] Obtaining multi-temporal optical remote sensing images of a target area, and reconstructing the multi-temporal optical remote sensing images to obtain high-resolution farmland images;

[0006] Performing feature extraction and crop identification on the high-resolution farmland images to obtain a crop type distribution map;

[0007] Performing multi-temporal comparison analysis on the crop type distribution map to obtain farmland utilization change data, and generating a farmland planting situation monitoring report based on the farmland utilization change data.

[0008] In combination with the first aspect, in a first implementation manner of the first aspect of the present application, the obtaining of the multi-temporal optical remote sensing images of the target area and the reconstruction of the multi-temporal optical remote sensing images to obtain the high-resolution farmland images comprises:

[0009] Obtaining multi-temporal optical remote sensing images of a target area;

[0010] 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 farmland sparse basis dictionary;

[0011] statistically analyze texture response patterns of different crop types in each spectral band in the multi-temporal optical remote sensing images, and construct a farmland texture correlation matrix based on the texture response patterns;

[0012] acquire time information of the multi-temporal optical remote sensing images and identify corresponding crop growth periods, calculate a crop phenology weight factor according to the crop growth periods, and determine a sparsity threshold value by using the crop phenology weight factor;

[0013] based on the farmland sparse basis dictionary, the farmland texture correlation matrix, and the sparsity threshold value, iteratively reconstruct a transmission channel of the multi-temporal optical remote sensing images to obtain a channel response matrix;

[0014] perform double-driven super-resolution reconstruction on the multi-temporal optical remote sensing images according to the channel response matrix to obtain a high-resolution farmland image.

[0015] In a second implementation manner of the first aspect, the double-driven super-resolution reconstruction on the multi-temporal optical remote sensing images according to the channel response matrix to obtain a high-resolution farmland image comprises:

[0016] a sparse reconstruction model is established by using the farmland sparse basis dictionary and the farmland texture correlation matrix, and an initial sparse processing result of the transmission channel of the multi-temporal optical remote sensing images is represented as sparse basis functions and sparse coefficients;

[0017] a candidate basis function is selected from the spatial basis functions of the farmland sparse basis dictionary one by one, a similarity between the candidate basis function and a known farmland plot shape is calculated, and a basis function with a similarity greater than a preset threshold value is selected as a structural constraint basis function;

[0018] the initial sparse processing result is reconstructed based on the sparse reconstruction model and the structural constraint basis function to generate a reconstruction result;

[0019] a candidate basis function with a minimum spectral angle distance is selected as an optimal basis function according to the reconstruction result and a characteristic spectral curve, and a residual value is updated;

[0020] the iteration number is controlled according to the sparsity threshold value, the candidate basis function that minimizes the residual value and meets the optimal basis function is repeatedly selected until the residual value converges or the sparsity threshold value is reached, and a channel response matrix is obtained.

[0021] In a third implementation manner of the first aspect, the double-driven super-resolution reconstruction on the multi-temporal optical remote sensing images according to the channel response matrix to obtain a high-resolution farmland image comprises:

[0022] input the channel response matrix and the multi-temporal optical remote sensing image into a double-driven super-resolution reconstruction network, perform multi-scale feature extraction through an encoder in the double-driven super-resolution reconstruction network to obtain an encoded feature map;

[0023] perform boundary keeping driving based on the encoded feature map, generate a boundary enhanced feature map, and perform spectral feature enhancement on the boundary enhanced feature map to obtain a spectral enhanced feature map;

[0024] perform up-sampling reconstruction on the spectral enhanced feature map through a decoder in the double-driven super-resolution reconstruction network, while keeping the plot boundary definition of the boundary enhanced feature map and the crop distinguishing feature of the spectral enhanced feature map, to obtain a high-resolution farmland image.

[0025] In a fourth implementation manner of the first aspect, the inputting of the channel response matrix and the multi-temporal optical remote sensing image into the double-driven super-resolution reconstruction network and the multi-scale feature extraction through the encoder in the double-driven super-resolution reconstruction network to obtain the encoded feature map include:

[0026] perform transmission distortion compensation on the multi-temporal optical remote sensing image by using the channel response matrix to obtain a channel corrected image;

[0027] input the channel corrected image into a feature extraction module of the encoder in the double-driven super-resolution reconstruction network to perform feature extraction to obtain original features of each scale;

[0028] perform channel attention processing on the original features of each scale to obtain channel enhanced features;

[0029] perform adaptive adjustment on the channel enhanced features based on quality evaluation information of the channel response matrix to output the encoded feature map.

[0030] In a fifth implementation manner of the first aspect, the feature extraction and crop recognition on the high-resolution farmland image to obtain a crop type distribution map include:

[0031] perform area statistical analysis on farmland plots in the high-resolution farmland image to obtain scale matching features;

[0032] input the scale matching features into a multi-scale feature pyramid for parallel processing to obtain multi-scale spatial features;

[0033] input the multi-scale spatial features into a time sequence attention module for time dimension fusion based on time sequence information of the multi-temporal optical remote sensing image to obtain time sequence enhanced features;

[0034] According to the timing enhancement feature, the geometric boundary information of the farmland plot and the crop phenology feature are combined for feature correction, and a farmland multi-scale feature map is obtained;

[0035] Based on the farmland multi-scale feature map, crop recognition is performed to obtain a crop type distribution map.

[0036] In combination with the first aspect, in a sixth implementation manner of the first aspect of the present application, the crop recognition based on the farmland multi-scale feature map to obtain the crop type distribution map comprises:

[0037] The farmland multi-scale feature map is input into a spectral feature separation layer in a crop recognition model to perform spectral component separation, and a spectral separation result is obtained;

[0038] Based on the grain crop component in the spectral separation result, crop feature enhancement is performed to obtain an enhanced spectral feature;

[0039] The enhanced spectral feature is input into a crop classification layer in the crop recognition model to perform type recognition, and a crop classification result is obtained;

[0040] Based on the crop classification result, connected region analysis is performed and combined with farmland plot boundary information to generate a crop type distribution map.

[0041] In combination with the first aspect, in a seventh implementation manner of the first aspect of the present application, the multi-temporal comparison analysis on the crop type distribution map to obtain the cultivated land use change data, and the generation of a cultivated land grain planting situation monitoring report based on the cultivated land use change data comprises:

[0042] The crop type distribution maps of different time phases are compared pixel by pixel according to the same spatial coordinates to identify plot change information;

[0043] According to the plot change information, a cultivated land use change mode is determined, and the occurrence area and spatial distribution of each cultivated land use change mode are calculated to obtain classification change data;

[0044] Based on the classification change data, the change time node of each changed plot is calculated, the month interval of change occurrence is determined by using the acquisition time of the multi-temporal optical remote sensing image, and the cultivated land use change data is established by combining the boundary coordinates and area information of the changed plot;

[0045] According to the cultivated land use change data, a key focus area with abnormal change frequency is identified, and change statistical data, trend analysis results and early warning area information are summarized to generate a cultivated land grain planting situation monitoring report.

[0046] Secondly, the present application provides a cultivated land grain planting situation dynamic monitoring system, which comprises:

[0047] An acquisition module is configured 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;

[0048] A crop recognition module is configured to perform feature extraction and crop recognition on the high-resolution farmland images to obtain a crop type distribution map;

[0049] A generation module is configured to perform multi-temporal comparison and analysis on the crop type distribution map to obtain farmland utilization change data, and generate a farmland planting situation monitoring report based on the farmland utilization change data.

[0050] The third aspect of the present application provides a computer readable storage medium, the computer readable storage medium stores instructions, when the instructions are run on a computer, the computer executes the above-mentioned farmland planting situation dynamic monitoring method.

[0051] In the technical solution provided by the present application, through the farmland block sparse channel estimation technology, the regular geometric distribution characteristics of the farmland blocks are fully utilized to construct a special sparse base dictionary, and the sparse reconstruction parameters are dynamically adjusted by combining the crop phenology weight factor, thereby effectively solving the quality degradation problem of satellite remote sensing images in the transmission process. The double-driven super-resolution reconstruction mechanism driven by the boundary preservation and the spectral feature enhancement is adopted to improve the image resolution while maintaining the clarity of the farmland block boundaries, and to highlight the spectral differentiation characteristics of food crops and economic crops. The farmland scale adaptive mechanism is adopted to dynamically adjust the convolution kernel size according to the block area, thereby effectively solving the technical problem that the traditional method cannot adapt to farmland blocks of different scales. The time sequence attention mechanism is used to fuse the multi-temporal crop growth characteristics, and the crop phenology priori knowledge is used to guide the time sequence weight distribution, thereby fully exploiting the crop growth time sequence rule. Compared with the static image analysis method, the present application has higher crop recognition accuracy. The present application adopts the spectral feature separation and crop feature enhancement technology to realize accurate differentiation of food crops such as rice, wheat and corn and economic crops, effectively identifies subtle "non-grain" changes, and solves the limitation that the traditional method can only detect obvious land use conversion. BRIEF DESCRIPTION OF DRAWINGS

[0052] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0053] Figure 1 The steps of the farmland planting situation dynamic monitoring method in the embodiments of the present application are shown in the following schematic diagram.

[0054] Figure 2A structure schematic diagram of a farmland plowing and planting condition dynamic monitoring system in the embodiment of the present application. DETAILED DESCRIPTION

[0055] The embodiment of the present application provides a farmland plowing and planting condition dynamic monitoring method, system and storage medium. The terms "first", "second", "third", "fourth" and the like (if any) in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and do not have to be used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged under appropriate circumstances, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "include" or "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product or device including a series of steps or units does not have to be limited to those steps or units clearly listed, but can include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0056] For the convenience of understanding, the specific flow of the embodiment of the present application is described below. Please refer to Figure 1 One embodiment of the farmland plowing and planting condition dynamic monitoring method in the embodiment of the present application comprises the following steps.

[0057] Step S1, obtaining multi-temporal optical remote sensing images of a target area, and reconstructing the multi-temporal optical remote sensing images to obtain high-resolution farmland images;

[0058] In this embodiment, multi-temporal optical remote sensing images are acquired at key phenological stages such as sowing period, seedling stage, jointing stage, heading stage, grain filling stage and maturity stage in the target area, and after geometric correction, radiation correction and atmospheric correction, all temporal data are registered in a unified coordinate system. Spatial feature analysis is performed on the remote sensing images, and the geometric boundary shape, spatial arrangement rule and structural characteristics of the farmland plot are extracted. These spatial basis functions are combined to form a sparse basis dictionary of farmland, which effectively expresses the prior knowledge of the block distribution of farmland. At the same time, the spectral texture response patterns of different crops in multiple bands such as red, green, blue, red edge, near-infrared and short-wave infrared in multi-temporal images are statistically analyzed, and the spatial and temporal variation characteristics are used to construct a farmland texture correlation matrix to reveal the spectral differences and spatial correlations of different crops at different stages. Combined with the image acquisition time, the corresponding crop growth period is identified, and the physiological change rule of the crop at each key stage is used to calculate the phenological weight factor, so that the reconstruction algorithm can adapt to the characteristics of crops at different stages, and the sparsity threshold is dynamically determined by the phenological weight factor. Based on the sparse basis dictionary of farmland, the farmland texture correlation matrix and the sparsity threshold, the degradation channel in the image transmission process is iteratively reconstructed, the channel response is gradually optimized, and the channel response matrix representing the real spectrum and spatial characteristics of crops is obtained. The channel response matrix is input into the double-driven super-resolution reconstruction network, the farmland boundary preservation drive is used to ensure the continuity and clarity of the plot boundary, and the non-food detection drive is used to strengthen the distinguishing features between food crops and non-food crops, and high-resolution farmland images are generated.

[0059] Step S2, feature extraction and crop recognition are performed on the high-resolution farmland image to obtain a crop type distribution map;

[0060] In this embodiment, the area of each farmland plot in the high-resolution farmland image is statistically analyzed, and the scale matching features are constructed by calculating the mean, standard deviation, and quantile of the area distribution. The scale matching features are input into a multi-scale feature pyramid structure, and the multi-scale spatial features containing detailed information and global semantics are extracted under the parallel processing of different scale convolution kernels, so as to balance the fine texture of small plots and the overall spatial pattern of large plots. Combined with the time series information of multi-temporal optical remote sensing images, the multi-scale spatial features are input into the time sequence attention module, the different temporal features are weighted and fused through the self-attention mechanism, the time sequence enhanced features with the dynamic change law of crop growth cycle are generated, and the model can identify the spectral reflection and spatial texture change of crops at different phenological stages. According to the time sequence enhanced features, combined with the geometric boundary information of farmland plots and the phenological characteristics of crops at different growth stages, the feature expression is corrected, the farmland boundary is kept clear, the crop class feature is more representative, and the farmland multi-scale feature map is formed. The crop recognition is carried out on the farmland multi-scale feature map based on the classification and segmentation model of deep learning, the main food crops such as rice, wheat and corn are distinguished from economic crops, woodland or abandoned land plots, and the crop type distribution map is generated at the plot level to reflect the crop planting pattern of the target area.

[0061] Step S3, multi-temporal comparison analysis is performed on the crop type distribution map to obtain the cultivated land use change data, and a cultivated land grain planting situation monitoring report is generated based on the cultivated land use change data.

[0062] In this embodiment, the crop type distribution maps of different time periods are compared pixel by pixel under the unified spatial coordinate system, the change information of the plots is identified through the difference detection at the pixel level, and it is judged whether a plot changes from food crops to economic crops, woodland, construction land or abandoned land at different monitoring time periods. According to the plot change information, the change mode of cultivated land use is classified and identified, a change mode set including grain-to-economy conversion, cultivated land abandonment, farmland-to-non-agriculture conversion, and farmland-to-forest conversion is established, and the occurrence area and spatial distribution of each type of change mode are calculated to form classified change data with spatial and temporal distribution characteristics. Based on the classified change data, combined with the acquisition time of multi-temporal optical remote sensing images, the change time node of each changed plot is calculated, the month interval of change occurrence is determined through time sequence analysis method, and the boundary coordinates, area information and change type of the plot are stored together to form structured cultivated land use change data. The change frequency in the region is statistically analyzed by using the cultivated land use change data, so as to identify the key attention areas with abnormal change frequency, and the change statistical results and trend analysis data are obtained through aggregation calculation to mark the warning areas of “non-grain” or “non-agriculture” risk. The change statistical data, trend analysis results and warning area information are summarized to generate a cultivated land grain planting situation monitoring report.

[0063] In a specific embodiment, the process of performing step S1 can specifically include the following steps:

[0064] Obtaining multi-temporal optical remote sensing images of the target area;

[0065] Analyzing the spatial basis functions of the farmland plots in the multi-temporal optical remote sensing images, and combining the spatial basis functions to form a sparse basis dictionary of farmland;

[0066] 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;

[0067] Obtaining time information of the multi-temporal optical remote sensing images and identifying the corresponding crop growth periods, calculating crop phenology weight factors according to the crop growth periods, and determining a sparsity threshold using the crop phenology weight factors;

[0068] Based on the sparse basis dictionary of farmland, the farmland texture correlation matrix, and the sparsity threshold, iteratively reconstructing the transmission channel of the multi-temporal optical remote sensing images to obtain a channel response matrix;

[0069] According to the channel response matrix, performing double-driven super-resolution reconstruction on the multi-temporal optical remote sensing images to obtain high-resolution farmland images.

[0070] In this embodiment, multi-temporal optical remote sensing images of target areas are obtained at key growth stages such as the seeding period, seedling period, jointing period, heading period, filling period and maturity period. The images include high-resolution satellite images, multispectral images and hyperspectral images, and are subjected to geometric correction, radiation correction and atmospheric correction to eliminate sensor noise, illumination changes and atmospheric scattering interference, and to ensure that images of different time phases and different sources can be registered in a unified coordinate system. The spatial geometric features of farmland plots in the remote sensing images are analyzed, the geometric shape, boundary orientation and regularized distribution characteristics of each plot are extracted, and a set of spatial basis functions is generated through elliptical fitting and rectangularization processing methods. The spatial basis function combination forms a sparse basis dictionary of farmland, which is used as a priori knowledge base for image sparse reconstruction, thereby ensuring the authenticity of farmland boundaries and spatial layout during image reconstruction. The texture response patterns of different crop types in multiple spectral bands are statistically modeled and analyzed, such as analyzing the differences in crop chlorophyll content in the red edge band, analyzing the crop biomass characteristics in the near-infrared band, and identifying the changes in crop water content in the short-wave infrared band. The cross-band texture response patterns are integrated to construct a farmland texture correlation matrix, which reflects the spectral consistency and differences of crop categories in the spatio-temporal evolution process, and provides spectral dimension constraints for sparse channel estimation. The corresponding crop growth period is identified by combining the acquisition time information of the remote sensing images, and the crop phenology weight factor is calculated according to the physiological characteristic changes of each growth period. The phenology weight factor is modeled using an exponential decay function, so that the weight is higher at key periods such as the heading period and maturity period, and the weight is relatively lower at stages such as the seedling period and filling period, thereby dynamically determining the sparsity threshold value through the weight adjustment mechanism. The sparse basis dictionary of farmland, the farmland texture correlation matrix and the dynamically adjusted sparsity threshold value are used to iteratively reconstruct the transmission channel of multi-temporal remote sensing images. The improved orthogonal matching pursuit algorithm is used to gradually optimize the channel estimation, and the basis function closest to the real farmland geometric shape and crop spectral curve is continuously selected in the iteration process. The iteration stops when the residual energy is reduced to a preset threshold or the maximum iteration number is reached, and the channel response matrix is obtained, which can represent the spatial structure and crop spectral characteristics of the real farmland. According to the channel response matrix, the remote sensing images are subjected to double-driven super-resolution reconstruction. The boundary preservation drive maintains the clarity and continuity of the farmland plot boundaries through the boundary preservation loss function, and the non-food detection drive enhances the spectral and spatial discrimination between food crops and economic crops, forest land and non-agricultural areas through the crop segmentation loss function. The network uses progressive upsampling combined with a multi-scale boundary refinement module to increase the resolution while maintaining the integrity of the farmland texture features, and gradually optimizes the boundary expression and crop recognition ability in multi-stage training, and outputs high-resolution farmland images.

[0071] In a specific embodiment, the process of performing the step of performing double-driven super-resolution reconstruction on the multi-temporal optical remote sensing image according to the channel response matrix to obtain the high-resolution farmland image can specifically include the following steps:

[0072] The sparse reconstruction model is established by using the farmland sparse base dictionary and the farmland texture correlation matrix, and the transmission channel of the multi-temporal optical remote sensing image is represented as an initial sparse processing result of sparse base functions and sparse coefficients.

[0073] The candidate base functions are selected one by one from the spatial base functions of the farmland sparse base dictionary, the similarity between the candidate base functions and the known farmland plot shape is calculated, and the base functions with a similarity greater than a preset threshold are selected as the structural constraint base functions.

[0074] The initial sparse processing result is reconstructed based on the sparse reconstruction model and the structural constraint base functions to generate a reconstruction result.

[0075] The candidate base function with the minimum spectral angle distance is selected as the optimal base function according to the reconstruction result and the characteristic spectral curve, and the residual value is updated.

[0076] The number of iterations is controlled according to the sparsity threshold, the candidate base function that minimizes the residual value and meets the optimal base function is repeatedly selected until the residual converges or the sparsity threshold is reached, and the channel response matrix is obtained.

[0077] In this embodiment, a sparse reconstruction model is established based on a sparse base dictionary of farmland and a farmland texture correlation matrix. The degradation form of multi-temporal optical remote sensing images in the transmission channel is characterized by the initial sparse processing result through the base function expansion and sparse coefficient. The initial sparse processing result is equivalent to projecting the complex signal degradation process to the space of the sparse base dictionary, so that the main structure and texture pattern of the original channel can be approximated by a limited number of base functions. A candidate base function is selected from the space base function of the sparse base dictionary one by one, and the similarity of each candidate base function to the known farmland plot geometry is calculated. The similarity is measured by a combination of shape context descriptor and Hausdorff distance, etc. When the candidate base function has high consistency with the plot boundary in spatial topology and geometric distribution, the similarity value is greater than the preset threshold. At this time, the selected base function is screened as a structure constraint base function. Only the base function that meets the spatial geometric consistency constraint will participate in the subsequent reconstruction calculation, avoiding the accumulation of reconstruction errors caused by the introduction of irrelevant base functions. Under the framework of the sparse reconstruction model, the structure constraint base function selected is used to gradually reconstruct the initial sparse processing result, and the reconstruction result of each iteration is compared with the characteristic spectral curve of different crop categories. The spectral angle distance is used as the discrimination index, and the spectral similarity between the spectral features of the reconstruction result and the standard crop spectral curve is measured by calculating the angle size. The candidate base function with the smallest spectral angle distance is selected as the optimal base function from the candidate base function set, and the optimal base function is used to update the current residual value. The residual value reflects the difference between the original observed signal and the current reconstructed signal. As the optimal base function is continuously selected and superimposed, the residual value gradually decreases. In order to control the process of iteration, a sparsity threshold is introduced as a stopping condition. In each iteration, the optimal base function that minimizes the residual value and meets the geometric and spectral double constraints is selected, and the reconstruction result and the residual are continuously updated until the residual value converges below the preset convergence threshold or the maximum number of iterations set by the sparsity threshold is reached. After the iteration is completed, the channel response matrix is obtained.

[0078] In a specific embodiment, the process of performing step of performing double-driven super-resolution reconstruction on the multi-temporal optical remote sensing images according to the channel response matrix to obtain a high-resolution farmland image can specifically include the following steps:

[0079] The channel response matrix and the multi-temporal optical remote sensing images are input into the double-driven super-resolution reconstruction network, and multi-scale feature extraction is performed through the encoder in the double-driven super-resolution reconstruction network to obtain an encoded feature map.

[0080] Boundary preservation driving is performed based on the encoded feature map to generate a boundary enhanced feature map, and spectral feature enhancement is performed on the boundary enhanced feature map to obtain a spectral enhanced feature map.

[0081] The decoder in the dual-driven super-resolution reconstruction network is used for up-sampling and reconstructing the spectral enhancement feature map, while the boundary enhancement feature map is kept clear and the crop distinguishing feature of the spectral enhancement feature map is kept, so as to obtain a high-resolution farmland image.

[0082] In the embodiment, the channel response matrix and the multi-temporal optical remote sensing image are jointly input into the dual-driven super-resolution reconstruction network, the multi-scale convolution structure is used for feature extraction in the front-end encoder part of the network, the residual dense block is combined with the channel attention mechanism, so that the low-layer convolution can capture the fine-grained texture feature, and the high-layer convolution can obtain the global semantic expression, to form the encoding feature map, and the spatial distribution information of the farmland plot and the crop spectral response information are reserved. The boundary keeping driver is executed on the basis of the encoding feature map, the improved edge detection operator is used to extract the farmland plot boundary, the gradient loss and the total variation regularization term are combined to enhance the edge feature, and the boundary enhancement feature map is generated, so that the weight of the plot boundary in the feature space is amplified. The spectral feature enhancement link is introduced on the basis of the boundary enhancement, the spectral characteristics of crops in the key wavebands such as red light, red edge, near-infrared and short-wave infrared are used to perform channel-by-channel weighting correction on the feature space, so that the difference between the food crops and the economic crops in the spectral dimension is more obvious, and the spectral enhancement feature map is generated. The spectral enhancement feature map is input into the decoder part for up-sampling and reconstruction, the decoder uses the progressive up-sampling combined with the sub-pixel convolution and the bilinear interpolation to gradually restore the spatial resolution, and the residual connection of the boundary enhancement feature is introduced at each up-sampling stage, so that the boundary clarity is continuously strengthened in the resolution restoration process, and the high-resolution farmland image is output.

[0083] In a specific embodiment, the step of inputting the channel response matrix and the multi-temporal optical remote sensing image into the dual-driven super-resolution reconstruction network and performing multi-scale feature extraction by the encoder in the dual-driven super-resolution reconstruction network can specifically include the following steps:

[0084] The multi-temporal optical remote sensing image is compensated for transmission distortion by using the channel response matrix, to obtain a channel corrected image;

[0085] The channel corrected image is input into the feature extraction module of the encoder in the dual-driven super-resolution reconstruction network for feature extraction, to obtain original features of each scale;

[0086] The channel attention processing is performed on the original features of each scale, to obtain channel enhanced features;

[0087] The channel enhanced features are adaptively adjusted based on the quality evaluation information of the channel response matrix, to output the encoding feature map.

[0088] In this embodiment, the channel response matrix is used to compensate for the transmission distortion of multi-temporal optical remote sensing images. The channel response matrix reflects the real transmission characteristics and degradation mode of remote sensing signals at different wavebands. Through the pixel-by-pixel convolution operation or deconvolution operation of the channel response matrix and the image data, the compensation of the blur, noise and offset caused by transmission distortion in the image is realized, and the channel corrected image is obtained. The channel corrected image is input into the encoder part of the dual-driven super-resolution reconstruction network. The feature extraction module of the encoder is composed of multi-scale convolution layers and residual dense blocks, which can capture low-level texture details and high-level spatial structure information in the image at the same time, and obtain original features at each scale through hierarchical convolution extraction. Channel attention processing is performed on the original features at each scale. The channel attention mechanism obtains the importance distribution of different channels through global average pooling and weighted calculation of each feature channel, and generates weight coefficients using a fully connected layer and an activation function to highlight the channels related to crop identification and boundary preservation in the feature map and suppress background noise or non-agricultural land interference. At the same time, considering the quality difference of the channel response matrix itself, the channel enhanced features are adaptively adjusted based on the quality evaluation information of the channel response matrix, which includes mean square error, correlation coefficient and spectral similarity. The reliability of image correction in a certain part is judged through these indicators. If the channel quality evaluation is excellent, the channel enhanced features remain the original weight unchanged; if the evaluation is good, further smoothing processing is performed on the basis of the channel attention output to reduce the noise sensitivity; if the evaluation is medium or poor, regularization constraint is added on the feature channel or redundancy correction is introduced in the decoding link. Through the adaptive adjustment process, the encoder can output boundary-sensitive geometric features and crop classification-effective spectral features, and can automatically optimize the feature weight according to the difference of channel quality, avoiding the decline of overall recognition accuracy caused by local channel distortion.

[0089] In a specific embodiment, the process of performing step S2 can specifically include the following steps:

[0090] Performing area statistical analysis on the farmland plots in the high-resolution farmland image to obtain scale matching features;

[0091] Inputting the scale matching features into a multi-scale feature pyramid for parallel processing to obtain multi-scale spatial features;

[0092] Based on the time series information of the multi-temporal optical remote sensing image, inputting the multi-scale spatial features into a time sequence attention module for time dimension fusion to obtain time sequence enhanced features;

[0093] According to the time sequence enhanced features, combining the geometric boundary information of the farmland plots and the crop phenology features to perform feature correction to obtain farmland multi-scale feature maps;

[0094] Performing crop identification based on the farmland multi-scale feature maps to obtain a crop type distribution map.

[0095] In this embodiment, the area of each field block in the high-resolution field image is counted and analyzed, the pixel number and the actual area of each block are calculated, the statistical quantities such as the mean, variance and quantile of the overall block area are obtained, and the scale matching feature is generated therefrom. The scale matching feature reflects the spatial distribution law and area difference of the target region field, so that the feature extraction process can use adaptive convolution kernel size and feature fusion strategy according to field blocks of different scales, thereby avoiding the problems of loss of small block details or insufficient features of large blocks. The scale matching feature is input into a multi-scale feature pyramid structure for parallel processing. The multi-scale feature pyramid structure sets multiple scale layers, each scale layer uses different size of convolution kernel and number of feature channels to extract image features, so that the network can capture local texture details and global spatial semantic information of the block at the same level, and obtain multi-scale spatial features. In combination with the time sequence information of multi-temporal optical remote sensing images, the multi-scale spatial features are input into a time sequence attention module, and the self-attention mechanism is used to assign weights to features of different time phases, highlight the stages with obvious time sequence differences such as sowing period and mature period, and reduce the weight of stages with small spectral differences such as seedling period and grain filling period, so as to enhance the features of the crop growth cycle in the time dimension, and obtain time sequence enhanced features. According to the time sequence enhanced features, the features are corrected in combination with the geometric boundary information of the field block and the crop phenology features. The geometric boundary information ensures that the field boundary is clear and continuous in the feature expression process, and avoids the boundary blur caused by up-sampling or convolution operation. The crop phenology features dynamically weight the spectral response of different crops in each growth period, so that the feature expression is more in line with the real physiological changes of crops, thereby generating a field multi-scale feature map. Based on the field multi-scale feature map, crop recognition is performed, the classification and segmentation model of deep learning is used to distinguish the crop type of each block, the main food crops such as rice, wheat and corn are distinguished from economic crops, woodland or abandoned land blocks, and a crop type distribution map is generated at the block level.

[0096] In a specific embodiment, the process of performing crop recognition based on the field multi-scale feature map to obtain the crop type distribution map can specifically include the following steps:

[0097] The field multi-scale feature map is input into a spectral feature separation layer in the crop recognition model to separate the spectral components, and a spectral separation result is obtained;

[0098] Based on the grain crop component in the spectral separation result, crop feature enhancement is performed to obtain enhanced spectral features;

[0099] The enhanced spectral features are input into a crop classification layer in the crop recognition model for type recognition, and a crop classification result is obtained;

[0100] Based on the crop classification result, a connected region analysis is performed and combined with farmland plot boundary information to generate a crop type distribution map.

[0101] In this embodiment, the farmland multi-scale feature map is input into a spectral feature separation layer in the crop recognition model. The spectral feature separation layer separates the spectral information in the input features according to the waveband response mode of different crops by independent component analysis or non-negative matrix factorization algorithm, to obtain a spectral separation result. In this process, the complex mixed spectral signal is divided into subspaces such as grain crop component, economic crop component and background component, to ensure that the features of different crops in the spectral dimension can be expressed separately and the class uncertainty caused by mixed pixels is avoided. The grain crop component is extracted from the spectral separation result, and crop feature enhancement is performed on the grain crop component. The enhancement module amplifies the key spectral response of the grain crops in the red edge band, near-infrared band and short-wave infrared band by weighting, so that the main grain crops such as rice, wheat and corn are more prominent in the feature space, and the interference of economic crops and non-agricultural areas in these key bands is suppressed, to obtain enhanced spectral features. The enhanced spectral features are input into the crop classification layer in the crop recognition model. The classification layer is composed of a convolutional neural network or a deep residual network. The local pattern is extracted by convolution operation, and the class probability is calculated by a fully connected layer or a softmax function, to output the crop classification result. The crop recognition model can give a prediction label belonging to a certain crop category for each pixel or plot unit, and fully utilize the discrimination degree improvement brought by spectral enhancement in the class discrimination process, to improve the classification accuracy of grain crops and non-grain crops. Based on the crop classification result, a connected region analysis is performed to identify the spatial aggregation features in the classification result, to avoid the false classification of scattered small regions caused by local noise. Through the connectivity analysis, the pixels belonging to the same category and spatially adjacent are merged into a unified region, and the spatial correction is performed combined with the farmland plot boundary information, to correct the misclassified pixels near the boundary while maintaining the integrity of the plot boundary, to generate a crop type distribution map.

[0102] In a specific embodiment, the process of step S3 can specifically include the following steps:

[0103] The crop type distribution maps of different time phases are compared pixel by pixel according to the same spatial coordinates to identify plot change information;

[0104] According to the plot change information, a cultivated land use change mode is determined, and the occurrence area and spatial distribution of each cultivated land use change mode are calculated to obtain classification change data;

[0105] Based on the classification change data, the change time node of each changed plot is calculated, the month interval of the change is determined by the acquisition time of the multi-temporal optical remote sensing image, and the cultivated land use change data is established combined with the boundary coordinates and area information of the changed plot.

[0106] According to the change frequency anomaly of the cultivated land utilization change data, the change statistical data, the trend analysis result and the early warning area information are summarized, and a cultivated land planting situation monitoring report is generated.

[0107] In the embodiment, the crop type distribution maps of different time phases are registered in a unified spatial coordinate system. The category labels of each pixel are detected for difference. If the category of a pixel changes from grain crop to economic crop, woodland, construction land or abandoned state in two or more time phases, it is determined that the corresponding pixel has a land use change, and the land change information is identified. According to the land change information, the change mode of the cultivated land use is determined. Different change modes include grain-to-economic conversion, agricultural-to-non-agricultural conversion, agricultural-to-forest conversion and farmland abandonment, etc. The pixel of each change mode is aggregated and counted to calculate the occurrence area and spatial distribution of different modes in the region, and the classification change data with classification and spatial-temporal attributes are obtained to reflect the dynamic pattern of the cultivated land use in different time stages. Based on the classification change data, the change time node corresponding to each changed land is calculated. By comparing the category differences of different time phases, the start and end time of the change are determined. The acquisition time of the multi-temporal optical remote sensing image is used as the time reference to locate the change event to a specific month interval. Combined with the boundary coordinates and area information of the changed land, a cultivated land use change data file is established for each changed land. The file contains land boundary shape, area size, change category and time interval, etc. According to the cultivated land use change data, the overall time sequence statistics and frequency analysis of the region are performed to identify the key attention areas with multiple changes in a short time or abnormally high change proportion. All the cultivated land use change data are summarized to generate a comprehensive output including change statistical results, trend analysis results and early warning area information, and a cultivated land planting situation monitoring report is formed.

[0108] The above describes the cultivated land planting situation dynamic monitoring method in the embodiment of the application. The following describes the cultivated land planting situation dynamic monitoring system in the embodiment of the application. Please refer to Figure 2 An embodiment of the cultivated land planting situation dynamic monitoring system in the embodiment of the application includes:

[0109] The acquisition module is configured to acquire multi-temporal optical remote sensing images of a target region, and reconstruct the multi-temporal optical remote sensing images to obtain high-resolution farmland images.

[0110] The crop recognition module is configured to perform feature extraction and crop recognition on the high-resolution farmland images to obtain a crop type distribution map.

[0111] The generating module is used for performing multi-time phase comparison analysis on the crop type distribution map, obtaining cultivated land utilization change data, and generating a cultivated land grain planting situation monitoring report based on the cultivated land utilization change data.

[0112] Through the cooperation of the above-mentioned components, through the farmland block sparse channel estimation technology, the regular geometric distribution characteristics of the farmland block are fully utilized to construct a special sparse base dictionary, the crop phenology weight factor is combined to dynamically adjust the sparse reconstruction parameter, the quality degradation problem of satellite remote sensing image in the transmission process is effectively solved, the double-driven super-resolution reconstruction mechanism of boundary preservation driving and spectral feature enhancement driving is adopted, the image resolution is improved while the farmland block boundary definition is maintained, and the spectral distinguishing features of food crops and economic crops are highlighted, the farmland scale adaptive mechanism is adopted to dynamically adjust the convolution kernel size according to the block area, the technical difficulty that the traditional method cannot adapt to different scale farmland blocks is effectively solved, the time sequence attention mechanism is used to fuse the multi-time phase crop growth characteristics, the crop phenology priori knowledge is combined to guide the time sequence weight distribution, the crop growth time sequence rule is fully tapped, and compared with the static image analysis method, the crop recognition precision is higher. The spectral feature separation and crop feature enhancement technology are adopted, the accurate distinction of rice, wheat, corn and other food crops and economic crops is realized, the subtle "non-grain" change is effectively identified, and the limitation that the traditional method can only detect obvious land use conversion is solved.

[0113] An embodiment of the present application further provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to realize the above method. It can be understood that the computer readable storage medium in the embodiment can be a volatile readable storage medium or a non-volatile readable storage medium.

[0114] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer readable storage medium, and when executed, can include the processes of the above-mentioned embodiment methods. Any reference to memory, storage, database or other medium provided by the present application and used in the 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. As an illustration but not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM, etc.

[0115] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-mentioned system, system and unit can refer to the corresponding processes in the foregoing method embodiments, which will not be described here.

[0116] The integrated unit, if implemented in the form of a software function unit and sold or used as an independent product, can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application essentially or the part that contributes to the prior art or the whole or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium, and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present application. The foregoing storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various program code storage media.

[0117] The above-described embodiments are only used to illustrate the technical solutions of the present application, and are not intended to limit the present application; although the present application has been described in detail with reference to the foregoing embodiments, it should be understood by those skilled in the art that the technical solutions recorded in the foregoing embodiments can still be modified, or some technical features can be replaced by equivalent replacements; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for dynamically monitoring the situation of ploughing and planting, characterized in that, The method comprises the following steps: acquiring multi-temporal optical remote sensing images of a target area, and reconstructing the multi-temporal optical remote sensing images to obtain high-resolution farmland images; specifically, acquiring multi-temporal optical remote sensing images of a target area; analyzing the spatial base functions of farmland plots in the multi-temporal optical remote sensing images, combining the spatial base functions to form a sparse base dictionary of farmland, and calculating the texture response mode 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 mode; acquiring time information of the multi-temporal optical remote sensing images and identifying the corresponding crop growth period, calculating the crop phenology weight factor according to the crop growth period, and determining the sparsity threshold value by using the crop phenology weight factor; based on the sparse base dictionary of farmland, the farmland texture correlation matrix and the sparsity threshold value, iteratively reconstructing the transmission channel of the multi-temporal optical remote sensing images to obtain a channel response matrix; using the sparse base dictionary of farmland and the farmland texture correlation matrix to establish a sparse reconstruction model, and expressing the transmission channel of the multi-temporal optical remote sensing images as an initial sparse processing result of sparse base functions and sparsity; selecting candidate base functions one by one from the spatial base functions of the sparse base dictionary of farmland, calculating the similarity of the candidate base functions and the known farmland plot shape, and screening out the base functions with a similarity greater than a preset threshold value as structure constraint base functions; reconstructing the initial sparse processing result based on the sparse reconstruction model and the structure constraint base functions to generate a reconstruction result; selecting the candidate base function with the minimum spectral angle distance as the optimal base function according to the reconstruction result and the characteristic spectral curve, and updating the residual value, which reflects the difference between the original observation signal and the current reconstructed signal; controlling the number of iterations according to the sparsity threshold value, repeatedly selecting the candidate base function that minimizes the residual value and meets the optimal base function, until the residual converges or the sparsity threshold value is reached, to obtain a channel response matrix; performing feature extraction and crop recognition on the high-resolution farmland images to obtain a crop type distribution map; performing multi-temporal comparative analysis on the crop type distribution map to obtain cultivated land use change data, and generating a cultivated land grain planting situation monitoring report based on the cultivated land use change data.

2. The method of claim 1, wherein the method further comprises: The double-driven super-resolution reconstruction of the multi-temporal optical remote sensing images based on the channel response matrix comprises: inputting the channel response matrix and the multi-temporal optical remote sensing images into a double-driven super-resolution reconstruction network, performing multi-scale feature extraction through an encoder in the double-driven super-resolution reconstruction network to obtain an encoded feature map; performing boundary preservation driving based on the encoded feature map to generate a boundary enhanced feature map, and performing spectral feature enhancement on the boundary enhanced feature map to obtain a spectral enhanced feature map; performing up-sampling reconstruction on the spectral enhanced feature map through a decoder in the double-driven super-resolution reconstruction network, while maintaining the plot boundary definition of the boundary enhanced feature map and the crop distinguishing feature of the spectral enhanced feature map, to obtain high-resolution farmland images.

3. The method according to claim 2, wherein The channel response matrix and the multi-temporal optical remote sensing image are input into a double-driven super-resolution reconstruction network, multi-scale feature extraction is performed through an encoder in the double-driven super-resolution reconstruction network, and encoded feature maps are obtained. Transmission distortion compensation is performed on the multi-temporal optical remote sensing image by using the channel response matrix, and channel corrected images are obtained. The channel corrected images are input into a feature extraction module of the encoder in the double-driven super-resolution reconstruction network for feature extraction, and original features of each scale are obtained. Channel attention processing is performed on the original features of each scale, and channel enhanced features are obtained. The channel enhanced features are adaptively adjusted based on quality evaluation information of the channel response matrix, and encoded feature maps are output.

4. The method of claim 1, wherein the method further comprises: The high-resolution farmland image is input into the feature extraction module of the encoder in the double-driven super-resolution reconstruction network for feature extraction, and original features of each scale are obtained. The channel response matrix and the multi-temporal optical remote sensing image are input into a double-driven super-resolution reconstruction network, multi-scale feature extraction is performed through an encoder in the double-driven super-resolution reconstruction network, and encoded feature maps are obtained. Transmission distortion compensation is performed on the multi-temporal optical remote sensing image by using the channel response matrix, and channel corrected images are obtained. The channel corrected images are input into a feature extraction module of the encoder in the double-driven super-resolution reconstruction network for feature extraction, and original features of each scale are obtained. Channel attention processing is performed on the original features of each scale, and channel enhanced features are obtained. The channel enhanced features are adaptively adjusted based on quality evaluation information of the channel response matrix, and encoded feature maps are output.

5. The method according to claim 4, wherein, The high-resolution farmland image is input into the feature extraction module of the encoder in the double-driven super-resolution reconstruction network for feature extraction, and original features of each scale are obtained. The channel response matrix and the multi-temporal optical remote sensing image are input into a double-driven super-resolution reconstruction network, multi-scale feature extraction is performed through an encoder in the double-driven super-resolution reconstruction network, and encoded feature maps are obtained. Transmission distortion compensation is performed on the multi-temporal optical remote sensing image by using the channel response matrix, and channel corrected images are obtained. The channel corrected images are input into a feature extraction module of the encoder in the double-driven super-resolution reconstruction network for feature extraction, and original features of each scale are obtained. Channel attention processing is performed on the original features of each scale, and channel enhanced features are obtained.

6. The method of claim 1, wherein the method further comprises: The channel enhanced features are adaptively adjusted based on quality evaluation information of the channel response matrix, and encoded feature maps are output. The high-resolution farmland image is input into the feature extraction module of the encoder in the double-driven super-resolution reconstruction network for feature extraction, and original features of each scale are obtained. The channel response matrix and the multi-temporal optical remote sensing image are input into a double-driven super-resolution reconstruction network, multi-scale feature extraction is performed through an encoder in the double-driven super-resolution reconstruction network, and encoded feature maps are obtained. Transmission distortion compensation is performed on the multi-temporal optical remote sensing image by using the channel response matrix, and channel corrected images are obtained. The channel corrected images are input into a feature extraction module of the encoder in the double-driven super-resolution reconstruction network for feature extraction, and original features of each scale are obtained. Channel attention processing is performed on the original features of each scale, and channel enhanced features are obtained. The channel enhanced features are adaptively adjusted based on quality evaluation information of the channel response matrix, and encoded feature maps are output. The high-resolution farmland image is input into the feature extraction module of the encoder in the double-driven super-resolution reconstruction network for feature extraction, and original features of each scale are obtained. The channel response matrix and the multi-temporal optical remote sensing image are input into a double-driven super-resolution reconstruction network, multi-scale feature extraction is performed through an encoder in the double-driven super-resolution reconstruction network, and encoded feature maps are obtained. Transmission distortion compensation is performed on the multi-temporal optical remote sensing image by using the channel response matrix, and channel corrected images are obtained. The channel corrected images are input into a feature extraction module of the encoder in the double-driven super-resolution reconstruction network for feature extraction, and original features of each scale are obtained. Channel attention processing is performed on the original features of each scale, and channel enhanced features are obtained. The channel enhanced features are adaptively adjusted based on quality evaluation information of the channel response matrix, and encoded feature maps are output. The high-resolution farmland image is input into the feature extraction module of the encoder in the double-driven super-resolution reconstruction network for feature extraction, and original features of each scale are obtained.

7. A system for dynamically monitoring the plowing and planting of fields, characterized by, A method for performing the method of claim 1-6, comprising: an acquisition module configured 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 comprising: acquiring multi-temporal optical remote sensing images of a target area; analyzing the spatial basis functions of farmland plots in the multi-temporal optical remote sensing images, combining the spatial basis functions to form a sparse basis dictionary of farmland; counting 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; obtaining time information of the multi-temporal optical remote sensing images and identifying the corresponding crop growth period, calculating the crop phenology weight factor according to the crop growth period, and determining the sparsity threshold using the crop phenology weight factor; based on the sparse basis dictionary of farmland, the farmland texture correlation matrix and the sparsity threshold, iteratively reconstructing the transmission channel of the multi-temporal optical remote sensing images to obtain a channel response matrix; using the sparse basis dictionary of farmland and the farmland texture correlation matrix to establish a sparse reconstruction model, and expressing the transmission channel of the multi-temporal optical remote sensing images as an initial sparse processing result of sparse basis functions and sparsity; selecting candidate basis functions one by one from the spatial basis functions of the sparse basis dictionary of farmland, calculating the similarity of the candidate basis functions with known farmland plot shapes, and selecting basis functions with a similarity greater than a preset threshold as structure constraint basis functions; reconstructing the initial sparse processing result based on the sparse reconstruction model and the structure constraint basis functions to generate a reconstruction result; selecting the candidate basis function with the smallest spectral angle distance as the optimal basis function according to the reconstruction result and the characteristic spectral curve, and updating the residual value, which reflects the difference between the original observation signal and the current reconstructed signal; controlling the number of iterations according to the sparsity threshold, repeatedly selecting the candidate basis function that minimizes the residual value and meets the optimal basis function, until the residual converges or the sparsity threshold is reached, to obtain a channel response matrix; a crop identification module configured to perform feature extraction and crop identification on the high-resolution farmland images to obtain a crop type distribution map; a generation module configured to perform multi-temporal comparative analysis on the crop type distribution map to obtain farmland utilization change data, and generate a farmland cultivation situation monitoring report based on the farmland utilization change data.

8. A computer-readable storage medium, characterized in that, A computer program is stored thereon, which, when executed by a processor, causes the processor to perform the method of claim 1-6.

Images