Blurred image optimization enhancement system based on artificial intelligence

Abstract

The invention discloses a blurred image optimization enhancement system based on artificial intelligence, which relates to the field of blurred image optimization enhancement and comprises a data acquisition module, a primary processing module and a deep processing module. The method comprises the following steps: analyzing dynamic blurring, out-of-focus blurring and atmosphere blurring of a crop remote sensing image to obtain a dynamic blurring characteristic value of the crop remote sensing image, an out-of-focus blurring characteristic value of the crop remote sensing image and an atmosphere blurring characteristic value of the crop remote sensing image; the method comprises the following steps: respectively judging dynamic blurring, defocus blurring and atmosphere blurring states of a crop remote sensing image, carrying out corresponding optimization adjustment, judging a comprehensive pathological state of a crop to obtain a crop pathological heterogeneity value of each region, and carrying out grading processing on each region of the crop remote sensing image to obtain a crop pathological heterogeneity value of each region; and corresponding optimization enhancement measures are matched, so that interference of various factors in the acquisition process of the crop remote sensing image can be comprehensively considered, and the accuracy of crop growth monitoring is ensured.

Description

Technical Field

[0001] This invention relates to the field of blurry image optimization and enhancement, specifically to a blurry image optimization and enhancement system based on artificial intelligence. Background Technology

[0002] With the continuous development of modern agriculture, crop health monitoring is crucial for ensuring food quality and yield. However, traditional methods relying too heavily on manual inspections are not only inefficient but also susceptible to various subjective factors, making it difficult to accurately and comprehensively grasp the growth status of crops. As agriculture becomes increasingly refined and large-scale, artificial intelligence-based fuzzy image optimization and enhancement systems have emerged.

[0003] Meanwhile, the acquisition of remote sensing images of crops is easily affected by various factors, such as atmospheric scattering, camera movement, and lens inaccuracy. Existing AI-based blurry image optimization and enhancement systems cannot accurately identify and adjust the above blurry states in a timely manner, nor can they perform multi-dimensional analysis of crop pathological states, nor can they match corresponding image optimization and enhancement measures for different pathological risk areas. This results in the loss of details and low clarity in the acquired remote sensing images of crops, inaccurate monitoring of crop growth, delays in the prevention and control of crop diseases and pests, and increases the difficulty and cost of prevention and control.

[0004] To address the aforementioned shortcomings, a technical solution is provided. Summary of the Invention

[0005] To address the technical problems raised in the background section, this invention is proposed. Embodiments of this invention provide an artificial intelligence-based blurry image optimization and enhancement system.

[0006] The objective of this invention can be achieved through the following technical solution: an artificial intelligence-based fuzzy image optimization and enhancement system, comprising a data acquisition module, a preliminary processing module, and a depth processing module.

[0007] The preliminary processing module includes a fuzzy analysis unit and an optimization adjustment unit. The fuzzy analysis unit receives frequency domain information, grayscale information, edge information, color channel information, and wavelet transform information to analyze the dynamic blur, defocus blur, and atmospheric blur of the crop remote sensing image, so as to obtain the dynamic blur feature value, defocus blur feature value, and atmospheric blur feature value of the crop remote sensing image. The optimization adjustment unit determines the dynamic blur, defocus blur, and atmospheric blur states of the crop remote sensing image respectively and performs corresponding optimization adjustments.

[0008] The deep processing module includes a pathological analysis unit and an optimization and enhancement unit. The pathological analysis unit is used to receive leaf skeleton information and canopy information from remote sensing images of crops in each region, and to determine the comprehensive pathological state of crops to obtain pathological heterogeneity values ​​for each region. The optimization and enhancement unit is used to receive pathological heterogeneity values ​​for crops in each region, to perform hierarchical processing on each region of the remote sensing image of crops, and to match corresponding optimization and enhancement measures.

[0009] Furthermore, the corresponding optimization and adjustment analysis steps are as follows:

[0010] If the dynamic blur feature value of the crop remote sensing image is greater than the set dynamic blur feature threshold δ1, then the corresponding optimization and adjustment measure one is adopted; if the defocus blur feature value of the crop remote sensing image is greater than the set defocus blur feature threshold δ2, then the corresponding optimization and adjustment measure two is adopted; if the atmospheric blur feature value of the crop remote sensing image is greater than the set atmospheric blur feature threshold δ3, then the corresponding optimization and adjustment measure three is adopted.

[0011] Furthermore, the atmospheric blur feature value analysis steps for the remote sensing image of crops are as follows:

[0012] The color channel information of the frequency domain image of the crop remote sensing image is obtained. The pixel values ​​of all pixels in each channel are summed and then divided by the total number of pixels in each channel to obtain the average pixel value for each channel. The standard deviation of the average pixel value for each channel is also obtained and accumulated to obtain the image pass mean divergence of the crop remote sensing image. A two-dimensional wavelet transform is used to decompose the crop remote sensing image into vertical high-frequency sub-bands, diagonal high-frequency sub-bands, horizontal high-frequency sub-bands, and low-frequency sub-bands. The low-frequency sub-bands are then decomposed a second time to obtain new vertical high-frequency sub-bands, diagonal high-frequency sub-bands, horizontal high-frequency sub-bands, and low-frequency sub-bands. This process is repeated continuously. Repeat the above operations, increase the decomposition scale of the low-frequency sub-band, and calculate the sum of squares of the coefficients in the vertical high-frequency sub-band, diagonal high-frequency sub-band, and horizontal high-frequency sub-band at each scale. Summate these values ​​to obtain the detail energy values ​​at each scale. Compare these values ​​with the set threshold α4, and count the percentage of detail energy values ​​at each scale that are lower than the threshold α4. Mark this percentage as the multi-scale detail energy reduction rate of the crop remote sensing image. Normalize this rate with the image pass-through average dispersion of the crop remote sensing image. Divide the multi-scale detail energy reduction rate of the crop remote sensing image by the image pass-through average dispersion of the crop remote sensing image to obtain the atmospheric blur feature value of the crop remote sensing image.

[0013] Furthermore, the dynamic blur feature value analysis steps for the crop remote sensing image are as follows:

[0014] The remote sensing image of crops is converted to grayscale to obtain the grayscale values ​​at each coordinate point. These grayscale values ​​are labeled as image functions of the crop remote sensing image. Taking the center pixel of the image as the reference, the relative displacement of any other pixel point to the reference point is calculated. This process is repeated for all pixel combinations in the remote sensing image. The grayscale value of the reference point with the relative displacement in each group is multiplied by the grayscale value at the corresponding displacement and then summed to obtain grayscale similarity metrics under different relative displacements. Using θ as the direction, starting from the image center, the displacement is gradually increased along this direction, and the corresponding grayscale similarity metrics are calculated. These grayscale similarity metrics are then compared within the displacement range. Linear fitting is performed within the range to obtain the slope of the fitted line in each θ direction, and the absolute value is taken to obtain the decay gradient magnitude β of the gray-level similarity measure in each θ direction. The average value of the decay gradient magnitude of the gray-level similarity measure in each θ direction is counted and marked as τ. The number of directions in each θ direction where the decay gradient magnitude β of the gray-level similarity measure is less than 0.7×τ is counted and marked as the dynamic blur tendency value of the crop remote sensing image. The directional energy distribution width of the crop remote sensing image is weighted and calculated with the dynamic blur tendency value of the crop remote sensing image, and multiplied by the corresponding weight factor coefficient to obtain the dynamic blur feature value of the crop remote sensing image.

[0015] Furthermore, the directional energy spread width analysis steps for the crop remote sensing image are as follows:

[0016] A two-dimensional discrete Fourier transform is performed on the remote sensing image of crops to calculate the energy spectrum of the frequency domain image. The energy spectrum represents the energy distribution of each frequency component in the image. With the center of the remote sensing image of crops as the origin, straight lines are drawn in different directions θ, with the value of θ ranging from 0° to 180°. The frequency domain coordinates on the straight lines in each direction θ are calculated. A coordinate transformation is performed using a rotation matrix to map the coordinate points in the new coordinate system with θ as the direction. It is determined whether the coordinate points in the new coordinate system have an angular deviation of less than α1 in the direction with θ as the direction. If the above condition is met, the coordinate points are assigned to the set with θ as the direction. The energy values ​​in the set with θ as the direction are calculated and added to obtain the input energy value in the direction with θ as the direction. The maximum input energy value is obtained, and this direction is taken as the direction of the energy peak set of the remote sensing image. With the direction of the energy peak set of the remote sensing image as the center, the angle range corresponding to the input energy value decreasing to the α2 ratio is obtained and marked as the directional energy distribution width of the remote sensing image of crops.

[0017] Furthermore, the defocus blur feature value analysis steps for the remote sensing image of crops are as follows:

[0018] By plotting the spectral energy curve of the frequency domain image of a crop remote sensing image, tangent lines are drawn at each coordinate point of the curve, and the slope of each tangent line is obtained and marked as the energy drop amplitude value of each coordinate point. The coordinate point where the energy drop amplitude value first appears to be less than a set threshold α3 is then identified, and the frequency value corresponding to this coordinate point is marked as the cutoff frequency value of the crop remote sensing image. Edges in the crop remote sensing image are detected using the Canny operator, resulting in an edge image. By fitting local edge curves, the curvature of each edge point is calculated, and the average value is calculated to obtain the edge curvature mean value of the crop remote sensing image. The edge image is then shaped... The morphological thinning operation is performed to thin to a single pixel width, and the edges are dilated. When the dilated edges cover all pixels of the original edges, the size of the structuring element at this point is obtained and marked as the edge topology equivalent of the crop remote sensing image. The cutoff frequency value, edge curvature mean value, and edge topology equivalent of the crop remote sensing image are normalized. The cutoff frequency value, edge topology equivalent, and edge curvature mean value of the crop remote sensing image are added together, and the reciprocal of the sum is calculated. Then, the sum is multiplied by the correction factor coefficient to obtain the defocus blur feature value of the crop remote sensing image.

[0019] Furthermore, the analysis steps for the optimization and enhancement measures corresponding to the matching are as follows:

[0020] The pathological heterogeneity values ​​of crops in each region are compared with the preset pathological warning value ranges ζ1, ζ2, and ζ3. If the pathological heterogeneity value of a region is within the pathological warning value range ζ1, then the region is classified as a high-risk region for crop disease outbreak, and optimization and enhancement measures are implemented accordingly. If the pathological heterogeneity value of a region is within the pathological warning value range ζ2, then the region is classified as a medium-risk region for crop disease outbreak, and optimization and enhancement measures are implemented accordingly. If the pathological heterogeneity value of a region is within the pathological warning value range ζ3, then the region is classified as a normal region for crop disease, and no corresponding action is taken.

[0021] Furthermore, the steps for analyzing the heterogeneity of crop pathological values ​​in each region are as follows:

[0022] The canopy volume and canopy permeability of crops in remote sensing images are acquired, and then subtracted from the predefined standard values ​​for canopy volume and canopy permeability, respectively. The absolute values ​​are then taken to obtain the canopy volume deviation and canopy permeability dispersion values ​​for each region, which are labeled as gzr and gcl, respectively. These values ​​are then normalized with the leaf fold partial parameter qsz and the leaf skeleton morphology heterogeneity index XYZ for each region, and substituted into the predefined formula. The crop pathological heterogeneity value NBL for each region was calculated, where j1, j2, j3 and j4 are the preset influencing factor coefficients of canopy volume deviation, canopy permeability dispersion, leaf folding deviation, and leaf skeleton morphology heterogeneity index for each region, respectively, and e is a natural constant.

[0023] Furthermore, the steps for analyzing the partial parametric properties of leaf folds in each region are as follows:

[0024] The height values ​​of each point on the surface of crop leaves in remote sensing images of each region are obtained, and the average value is calculated to obtain the average height reference of the leaves. The difference between each point and the height reference is calculated and the absolute value is accumulated to obtain the leaf surface fold dispersion value. The leaf surface fold dispersion value is subtracted from the set standard value of leaf surface fold dispersion and the absolute value is taken to obtain the leaf surface fold partial parameter qsz of each region.

[0025] Furthermore, the steps for analyzing the leaf skeleton morphology heterogeneity index in each region are as follows:

[0026] The remote sensing image of crops is divided into several regions. The crop leaf skeleton of each region is binarized to obtain a binarized leaf skeleton image. The bifurcation points of the leaf skeleton are identified, and the extension scale, bifurcation deviation angle, and branch topography of each bifurcation point are obtained and summed to obtain the morphological feature values ​​of the two branches, which are labeled as XT1 and XT2. They are substituted into the set formula model XTD=|XT1-XT2| / (XT1+XT2) to calculate the shape feature offset value XTD of each leaf skeleton bifurcation point. The shape feature offset value of each leaf skeleton bifurcation point is compared with the set shape feature offset value of the bifurcation point. The morphological bias at the bifurcation point of the leaf skeleton is compared with the morphological bias threshold ρ1. If the morphological bias at the bifurcation point is greater than the morphological bias threshold ρ1, then the bifurcation point is a significant morphological difference bifurcation point. If the morphological bias at the bifurcation point is equal to the morphological bias threshold ρ1, then the bifurcation point is a critical morphological bifurcation point. If the morphological bias at the bifurcation point is less than the morphological bias threshold ρ1, then the bifurcation point is a normal morphological bifurcation point. The number of significant morphological difference bifurcation points, critical morphological bifurcation points, and normal morphological bifurcation points in each region are counted and labeled as λ1, λ2, and λ3, respectively. These values ​​are then substituted into the set formula. The leaf skeleton morphology heterogeneity index XYZ for each region was calculated. th1 and th2 are preset influence factor coefficients, which are the sum of significant morphological difference bifurcation points and critical morphological bifurcation points and the number of conventional morphological bifurcation points, respectively.

[0027] Compared with the prior art, the present invention has the following beneficial effects:

[0028] 1. This invention analyzes the dynamic blur, defocus blur, and atmospheric blur of remote sensing images of crops to obtain the dynamic blur feature values, defocus blur feature values, and atmospheric blur feature values ​​of remote sensing images of crops. It determines the dynamic blur, defocus blur, and atmospheric blur states of remote sensing images of crops separately and makes corresponding optimization adjustments. It can comprehensively consider the interference of various factors during the acquisition of remote sensing images of crops, such as atmospheric scattering, camera movement, and lens inaccuracy, thus ensuring the detail and clarity of remote sensing images of crops.

[0029] 2. This invention determines the comprehensive pathological state of crops to obtain the pathological heterogeneity values ​​of crops in each region. It receives the pathological heterogeneity values ​​of crops in each region, performs hierarchical processing on each region of the remote sensing image of crops, and matches corresponding optimization and enhancement measures. It can perform multi-dimensional analysis of the pathological state of crops and match corresponding image optimization and enhancement measures for different pathological risk areas. This can ensure the accuracy of crop growth monitoring and the timeliness of crop pest and disease control, while reducing the difficulty and cost of prevention and control. Attached Figure Description

[0030] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. The following drawings are not drawn to scale according to the actual size, but are intended to show the main idea of ​​the present invention.

[0031] Figure 1 This is a system block diagram of the present invention. Detailed Implementation

[0032] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are also within the scope of protection of the present invention.

[0033] like Figure 1 As shown, the AI-based fuzzy image optimization and enhancement system includes a data acquisition module, a preliminary processing module, and a depth processing module.

[0034] The data acquisition module is used to collect frequency domain information, grayscale information, edge information, color channel information, wavelet transform information, leaf skeleton information, and canopy information of remote sensing images of crops, and send them to the preliminary processing module and the depth processing module.

[0035] The preliminary processing module includes a fuzzy analysis unit and an optimization and adjustment unit;

[0036] The fuzzy analysis unit receives frequency domain information, grayscale information, edge information, color channel information, and wavelet transform information to analyze the dynamic blur, defocus blur, and atmospheric blur of the crop remote sensing image, in order to obtain the dynamic blur feature value, defocus blur feature value, and atmospheric blur feature value of the crop remote sensing image, as follows:

[0037] A two-dimensional discrete Fourier transform is performed on the remote sensing image of crops to calculate the energy spectrum of the frequency domain image. The energy spectrum represents the energy distribution of each frequency component in the image. With the center of the remote sensing image of crops as the origin, straight lines are drawn in different directions θ, with the value of θ ranging from 0° to 180°. The frequency domain coordinates on the straight lines in each direction θ are calculated. The coordinates are transformed using a rotation matrix to obtain the coordinate points in the new coordinate system with θ as the direction. It is determined whether the coordinate points in the new coordinate system have an angular deviation of less than α1 in the direction with θ as the direction. If the above condition is met, the coordinate points are assigned to the set with θ as the direction. The energy values ​​in the set with θ as the direction are calculated and added to obtain the input energy value in the direction with θ as the direction. The maximum input energy value is obtained and the direction is taken as the direction of the energy peak set of the remote sensing image. With the direction of the energy peak set of the remote sensing image as the center, the angle range corresponding to the input energy value dropping to the α2 ratio is obtained and marked as the directional energy distribution width of the remote sensing image of crops.

[0038] The remote sensing image of crops is converted to grayscale to obtain the grayscale values ​​at each coordinate point. These grayscale values ​​are labeled as image functions of the crop remote sensing image. Taking the center pixel of the image as the reference, the relative displacement of any other pixel point to the reference point is calculated. This process is repeated for all pixel combinations in the remote sensing image. The grayscale value of the reference point with the relative displacement in each group is multiplied by the grayscale value at the corresponding displacement and then summed to obtain grayscale similarity metrics under different relative displacements. Using θ as the direction, starting from the image center, the displacement is gradually increased along this direction, and the corresponding grayscale similarity metrics are calculated. These grayscale similarity metrics are then compared within the displacement range. Linear fitting is performed within the range to obtain the slope of the fitted line in each θ direction, and the absolute value is taken to obtain the decay gradient magnitude β of the gray-level similarity measure in each θ direction. The average value of the decay gradient magnitude of the gray-level similarity measure in each θ direction is counted and marked as τ. The number of directions in each θ direction where the decay gradient magnitude β of the gray-level similarity measure is less than 0.7×τ is counted and marked as the dynamic blur tendency value of the crop remote sensing image. The directional energy spread width of the crop remote sensing image is weighted and calculated with the dynamic blur tendency value of the crop remote sensing image, and multiplied by the corresponding weight factor coefficient to obtain the dynamic blur feature value of the crop remote sensing image.

[0039] By plotting the spectral energy curve of the frequency domain image of a crop remote sensing image, tangent lines are drawn at each coordinate point of the curve, and the slope of each tangent line is obtained and marked as the energy drop amplitude value of each coordinate point. The coordinate point where the energy drop amplitude value first appears to be less than a set threshold α3 is then identified, and the frequency value corresponding to this coordinate point is marked as the cutoff frequency value of the crop remote sensing image. The threshold α3 is a value less than zero, specifically determined by professionals in this field. The Canny operator is used to detect edges in the crop remote sensing image, obtaining an edge image. By fitting local edge curves, the curvature of each edge point is calculated, and the average value is calculated to obtain the edge of the crop remote sensing image. The edge mean value is used to perform morphological thinning on the edge image, down to a single pixel width, and then dilates the edge. When the dilated edge covers all pixels of the original edge, the size of the structuring element at this point is obtained and marked as the edge topology equivalent of the crop remote sensing image. The cutoff frequency value, edge mean value, and edge topology equivalent of the crop remote sensing image are normalized. The cutoff frequency value, edge topology equivalent, and edge mean value of the crop remote sensing image are added together, and the reciprocal of the sum is calculated. Then, a correction factor coefficient is multiplied to obtain the defocus blur feature value of the crop remote sensing image.

[0040] The color channel information of the frequency domain image of the crop remote sensing image is obtained. The pixel values ​​of all pixels in each channel are summed and then divided by the total number of pixels in each channel to obtain the average pixel value for each channel (red, green, and blue). The standard deviation of the average pixel value for each channel is also obtained and summed to obtain the image pass mean divergence of the crop remote sensing image. A two-dimensional wavelet transform is used to decompose the crop remote sensing image into vertical high-frequency sub-bands, diagonal high-frequency sub-bands, horizontal high-frequency sub-bands, and low-frequency sub-bands. The low-frequency sub-band is then decomposed a second time to obtain new vertical high-frequency sub-bands, diagonal high-frequency sub-bands, horizontal high-frequency sub-bands, and low-frequency sub-bands. The above operation is repeated continuously to increase the decomposition scale of the low-frequency sub-band. The sum of squares of the coefficients in the vertical high-frequency sub-band, diagonal high-frequency sub-band, and horizontal high-frequency sub-band at each scale is calculated and summed to obtain the detail energy value at each scale. This value is compared with the set threshold α4, and the percentage of detail energy values ​​at each scale that are lower than the threshold α4 is counted and marked as the multi-scale detail energy reduction rate of the crop remote sensing image. This rate is then normalized with the image pass mean divergence of the crop remote sensing image. The multi-scale detail energy reduction rate of the crop remote sensing image is divided by the image pass mean divergence of the crop remote sensing image to obtain the atmospheric blur feature value of the crop remote sensing image.

[0041] The optimization and adjustment unit is used to determine the dynamic blur, defocus blur, and atmospheric blur states of remote sensing images of crops, and to perform corresponding optimization and adjustments, as follows:

[0042] If the motion blur feature value of the crop remote sensing image is greater than the set motion blur feature threshold δ1, the corresponding optimization adjustment is to match the parameters of the corresponding integer filter function according to the value of the motion blur feature value exceeding the threshold, effectively removing the motion blur. There is no corresponding operation for other values. If the defocus blur feature value of the crop remote sensing image is greater than the set defocus blur feature threshold δ2, the corresponding optimization adjustment is to use Wiener filtering deconvolution, increase the number of iterations, so as to more fully perform Wiener filtering deconvolution on the blurred image and gradually restore the clear details of the image. There is no corresponding operation for other values. If the atmospheric blur feature value of the crop remote sensing image is greater than the set atmospheric blur feature value threshold δ3, the corresponding optimization adjustment is to use ground meteorological station data to verify and calibrate the atmospheric correction model, and at the same time fuse high-resolution atmospheric composition data to refine the spatial resolution of atmospheric correction. There is no corresponding operation for other values.

[0043] The deep processing module includes a pathological analysis unit and an optimization and enhancement unit;

[0044] The pathological analysis unit receives leaf skeleton and canopy information from remote sensing images of crops in various regions, and determines the overall pathological state of the crops to obtain the pathological heterogeneity values ​​of crops in each region, as follows:

[0045] The remote sensing image of crops is divided into several regions. The crop leaf skeleton of each region is binarized to obtain a binarized leaf skeleton image. The bifurcation points of the leaf skeleton are identified, and the extension scale, bifurcation deviation angle, and branch topography of each bifurcation point are obtained and summed to obtain the morphological feature values ​​of the two branches, which are labeled as XT1 and XT2. They are substituted into the set formula model XTD=|XT1-XT2| / (XT1+XT2) to calculate the shape feature offset value XTD of each leaf skeleton bifurcation point. The shape feature offset value of each leaf skeleton bifurcation point is compared with the set shape feature offset value of the bifurcation point. The morphological bias at the bifurcation point of the leaf skeleton is compared with the morphological bias threshold ρ1. If the morphological bias at the bifurcation point is greater than the morphological bias threshold ρ1, then the bifurcation point is a significant morphological difference bifurcation point. If the morphological bias at the bifurcation point is equal to the morphological bias threshold ρ1, then the bifurcation point is a critical morphological bifurcation point. If the morphological bias at the bifurcation point is less than the morphological bias threshold ρ1, then the bifurcation point is a normal morphological bifurcation point. The number of significant morphological difference bifurcation points, critical morphological bifurcation points, and normal morphological bifurcation points in each region are counted and labeled as λ1, λ2, and λ3, respectively. These values ​​are then substituted into the set formula. The leaf skeleton morphology heterogeneity index XYZ for each region was calculated. th1 and th2 are the preset influence factor coefficients of the sum of significant morphological difference bifurcation points and critical morphological bifurcation points and the number of conventional morphological bifurcation points, respectively, with values ​​of 1.04 and 3.01.

[0046] It should be noted that the extension scale of a branch is the spatial distance from the bifurcation point along the direction of the leaf vein branch to the end of the branch; the bifurcation deviation angle of a branch is the angle formed between the branch and the vertical direction of the leaf at the bifurcation point; and the branch extension area is the area that the branch expands and occupies on the plane.

[0047] The height values ​​of each point on the surface of crop leaves in remote sensing images of each region are obtained, and the average value is calculated to obtain the average height benchmark of the leaves. The difference between each point and the height benchmark is calculated, and the absolute values ​​are accumulated to obtain the leaf fold dispersion value. This value is subtracted from the set standard value of leaf fold dispersion, and the absolute value is taken to obtain the leaf fold partial parameter qsz for each region. The crop canopy volume measurement and canopy permeability of the remote sensing images are obtained, and the absolute values ​​are subtracted from the set standard values ​​of canopy volume measurement and canopy permeability, respectively, to obtain the canopy volume deviation and canopy permeability dispersion values ​​for each region, which are labeled as gzr and gcl, respectively. These values ​​are normalized with the leaf fold partial parameter qsz and the leaf skeleton morphology heterogeneity index XYZ for each region, and then substituted into the set formula. The crop pathological heterogeneity value NBL for each region was calculated, where j1, j2, j3 and j4 are the preset influencing factor coefficients of canopy volume deviation, canopy permeability dispersion, leaf folding deviation, and leaf skeleton morphology heterogeneity index for each region, respectively, set by professionals in this field, and e is a natural constant with a value of 2.718.

[0048] It should be noted that canopy volume is the space occupied by the crop canopy, while canopy permeability is the proportion of air volume within the canopy.

[0049] The optimization and enhancement unit receives the heterogeneous values ​​of crop pathology in each region, performs hierarchical processing on each region of the crop remote sensing image, and matches corresponding optimization and enhancement measures, as follows:

[0050] The heterogeneous values ​​of crop pathology in each region are compared with the preset pathological warning ranges ζ1, ζ2, and ζ3. If the heterogeneous values ​​of crop pathology in a region are within the pathological warning range ζ1, then the region is classified as a high-risk region for crop disease outbreak. The optimization and enhancement measures are to use deep learning image restoration technology, such as an image restoration model based on generative adversarial networks, and to train the network with a large number of clear and blurry image pairs, so that the network learns the mapping relationship from blurry images to clear images. If the heterogeneous values ​​of crop pathology in a region are within the pathological warning range ζ2, then the region is classified as a medium-risk region for crop disease outbreak. The optimization and enhancement measures are to adjust the gray-level histogram of the crop image to make the gray-level distribution of the image more uniform, enhance the contrast of the image, and thus improve the clarity of the image. If the heterogeneous values ​​of crop pathology in a region are within the pathological warning range ζ3, then the region is classified as a normal region for crop disease, and no corresponding operation is performed.

[0051] The foregoing description is illustrative of the invention and should not be construed as limiting it. Although several exemplary embodiments of the invention have been described, those skilled in the art will readily understand that many modifications can be made to the exemplary embodiments without departing from the novel teachings and advantages of the invention. Therefore, all such modifications are intended to be included within the scope of the invention as defined in the claims. It should be understood that the foregoing description is illustrative of the invention and should not be construed as limiting it to the specific embodiments disclosed, and modifications to the disclosed embodiments and other embodiments are intended to be included within the scope of the appended claims. The invention is defined by the claims and their equivalents.

Claims

1. An AI-based blurry image optimization and enhancement system, comprising a data acquisition module, characterized in that, Also includes: The preliminary processing module includes a fuzzy analysis unit and an optimization and adjustment unit; The fuzzy analysis unit is used to receive frequency domain information, grayscale information, edge information, color channel information and wavelet transform information, and to analyze the dynamic fuzziness, defocus fuzziness and atmospheric fuzziness of the crop remote sensing image in order to obtain the dynamic fuzziness feature value, defocus fuzziness feature value and atmospheric fuzziness feature value of the crop remote sensing image. The optimization and adjustment unit is used to determine the dynamic blur, defocus blur, and atmospheric blur states of remote sensing images of crops, and to make corresponding optimization and adjustments. The deep processing module includes a pathological analysis unit and an optimization and enhancement unit. The pathological analysis unit is used to receive leaf skeleton information and canopy information from remote sensing images of crops in various regions, and to determine the comprehensive pathological state of crops in order to obtain the pathological heterogeneity value of crops in each region. The optimization and enhancement unit is used to receive the heterogeneous values ​​of crop pathology in each region, perform hierarchical processing on each region of the remote sensing image of crops, and match the corresponding optimization and enhancement measures.

2. The artificial intelligence-based fuzzy image optimization and enhancement system according to claim 1, characterized in that, The corresponding optimization and adjustment analysis steps are as follows: If the dynamic blur feature value of the crop remote sensing image is greater than the set dynamic blur feature threshold one, then the corresponding optimization and adjustment measure one is adopted; if the defocus blur feature value of the crop remote sensing image is greater than the set defocus blur feature threshold two, then the corresponding optimization and adjustment measure two is adopted; if the atmospheric blur feature value of the crop remote sensing image is greater than the set atmospheric blur feature value threshold three, then the corresponding optimization and adjustment measure three is adopted.

3. The artificial intelligence-based fuzzy image optimization and enhancement system according to claim 2, characterized in that, The atmospheric blur feature value analysis steps for the remote sensing image of crops are as follows: The color channel information of the frequency domain image of the crop remote sensing image is obtained. The pixel values ​​of all pixels in each channel are summed and then divided by the total number of pixels in each channel to obtain the average pixel value for each channel. The standard deviation of the average pixel value for each channel is also obtained and accumulated to obtain the image pass mean divergence of the crop remote sensing image. A two-dimensional wavelet transform is used to decompose the crop remote sensing image into vertical high-frequency sub-bands, diagonal high-frequency sub-bands, horizontal high-frequency sub-bands, and low-frequency sub-bands. The low-frequency sub-bands are then decomposed a second time to obtain new vertical high-frequency sub-bands, diagonal high-frequency sub-bands, horizontal high-frequency sub-bands, and low-frequency sub-bands. This process is repeated continuously. Repeat the above operations, increase the decomposition scale of the low-frequency sub-band, and calculate the sum of squares of the coefficients in the vertical high-frequency sub-band, diagonal high-frequency sub-band, and horizontal high-frequency sub-band at each scale. Summate these values ​​to obtain the detail energy values ​​at each scale. Compare these values ​​with the set threshold α4, and count the percentage of detail energy values ​​at each scale that are lower than the threshold α4. Mark this percentage as the multi-scale detail energy reduction rate of the crop remote sensing image. Normalize this rate with the image pass-through average dispersion of the crop remote sensing image. Divide the multi-scale detail energy reduction rate of the crop remote sensing image by the image pass-through average dispersion of the crop remote sensing image to obtain the atmospheric blur feature value of the crop remote sensing image.

4. The artificial intelligence-based fuzzy image optimization and enhancement system according to claim 2, characterized in that, The steps for analyzing the dynamic blur feature values ​​of the remote sensing images of crops are as follows: The remote sensing image of crops is converted to grayscale to obtain the grayscale values ​​at each coordinate point. These grayscale values ​​are labeled as image functions of the crop remote sensing image. Taking the center pixel of the image as the reference, the relative displacement of any other pixel point to the reference point is calculated. This process is repeated for all pixel combinations in the remote sensing image. The grayscale value of the reference point with the relative displacement in each group is multiplied by the grayscale value at the corresponding displacement and then summed to obtain grayscale similarity metrics under different relative displacements. Using θ as the direction, starting from the image center, the displacement is gradually increased along this direction, and the corresponding grayscale similarity metrics are calculated. These grayscale similarity metrics are then compared within the displacement range. Linear fitting is performed within the range to obtain the slope of the fitted line in each θ direction, and the absolute value is taken to obtain the decay gradient magnitude β of the gray-level similarity measure in each θ direction. The average value of the decay gradient magnitude of the gray-level similarity measure in each θ direction is counted and marked as τ. The number of directions in each θ direction where the decay gradient magnitude β of the gray-level similarity measure is less than 0.7×τ is counted and marked as the dynamic blur tendency value of the crop remote sensing image. The directional energy distribution width of the crop remote sensing image is weighted and calculated with the dynamic blur tendency value of the crop remote sensing image, and multiplied by the corresponding weight factor coefficient to obtain the dynamic blur feature value of the crop remote sensing image.

5. The artificial intelligence-based fuzzy image optimization and enhancement system according to claim 2, characterized in that, The steps for analyzing the directional energy spread width of the remote sensing image of crops are as follows: A two-dimensional discrete Fourier transform is performed on the remote sensing image of crops to calculate the energy spectrum of the frequency domain image. The energy spectrum represents the energy distribution of each frequency component in the image. With the center of the remote sensing image of crops as the origin, straight lines are drawn in different directions θ, with the value of θ ranging from 0° to 180°. The frequency domain coordinates on the straight lines in each direction θ are calculated. A coordinate transformation is performed using a rotation matrix to map the coordinate points in the new coordinate system with θ as the direction. It is determined whether the coordinate points in the new coordinate system have an angular deviation of less than α1 in the direction with θ as the direction. If the above condition is met, the coordinate points are assigned to the set with θ as the direction. The energy values ​​in the set with θ as the direction are calculated and added to obtain the input energy value in the direction with θ as the direction. The maximum input energy value is obtained, and this direction is taken as the direction of the energy peak set of the remote sensing image. With the direction of the energy peak set of the remote sensing image as the center, the angle range corresponding to the input energy value decreasing to the α2 ratio is obtained and marked as the directional energy distribution width of the remote sensing image of crops.

6. The artificial intelligence-based fuzzy image optimization and enhancement system according to claim 2, characterized in that, The steps for analyzing the defocus blur feature value of the remote sensing image of crops are as follows: By plotting the spectral energy curve of the frequency domain image of a crop remote sensing image, tangent lines are drawn at each coordinate point of the curve, and the slope of each tangent line is obtained and marked as the energy drop amplitude value of each coordinate point. The coordinate point where the energy drop amplitude value first appears to be less than a set threshold α3 is then identified, and the frequency value corresponding to this coordinate point is marked as the cutoff frequency value of the crop remote sensing image. Edges in the crop remote sensing image are detected using the Canny operator, resulting in an edge image. By fitting local edge curves, the curvature of each edge point is calculated, and the average value is calculated to obtain the edge curvature mean value of the crop remote sensing image. The edge image is then shaped... The morphological thinning operation is performed to thin to a single pixel width, and the edges are dilated. When the dilated edges cover all pixels of the original edges, the size of the structuring element at this point is obtained and marked as the edge topology equivalent of the crop remote sensing image. The cutoff frequency value, edge curvature mean value, and edge topology equivalent of the crop remote sensing image are normalized. The cutoff frequency value, edge topology equivalent, and edge curvature mean value of the crop remote sensing image are added together, and the reciprocal of the sum is calculated. Then, the sum is multiplied by the correction factor coefficient to obtain the defocus blur feature value of the crop remote sensing image.

7. The artificial intelligence-based fuzzy image optimization and enhancement system according to claim 1, characterized in that, The analysis steps for the optimization and enhancement measures corresponding to the matching are as follows: The pathological heterogeneity values ​​of crops in each region are compared with the preset pathological warning value ranges ζ1, ζ2, and ζ3. If the pathological heterogeneity value of a region is within the pathological warning value range ζ1, then the region is classified as a high-risk region for crop disease outbreak, and optimization and enhancement measures are implemented accordingly. If the pathological heterogeneity value of a region is within the pathological warning value range ζ2, then the region is classified as a medium-risk region for crop disease outbreak, and optimization and enhancement measures are implemented accordingly. If the pathological heterogeneity value of a region is within the pathological warning value range ζ3, then the region is classified as a normal region for crop disease, and no corresponding action is taken.

8. The artificial intelligence-based fuzzy image optimization and enhancement system according to claim 7, characterized in that, The steps for analyzing crop pathological heterogeneity values ​​in each region are as follows: The canopy volume and canopy permeability of crops in remote sensing images are acquired and subtracted from the set standard values ​​of canopy volume and canopy permeability, respectively. The absolute values ​​are then taken to obtain the canopy volume deviation and canopy permeability dispersion values ​​of each region. These values ​​are then processed and calculated with the leaf folding deviation parameter and the leaf skeleton morphology heterogeneity index of each region to obtain the crop pathological heterogeneity values ​​of each region.

9. The artificial intelligence-based fuzzy image optimization and enhancement system according to claim 8, characterized in that, The steps for analyzing the partial parametric properties of leaf folds in each region are as follows: The height values ​​of each point on the surface of crop leaves in remote sensing images of each region are obtained, and the average value is calculated to obtain the average height reference of the leaves. The difference between each point and the height reference is calculated and the absolute value is accumulated to obtain the leaf surface fold dispersion value. The leaf surface fold dispersion value is subtracted from the set standard value of leaf surface fold dispersion and the absolute value is taken to obtain the leaf surface fold partial parameter of each region.

10. The artificial intelligence-based fuzzy image optimization and enhancement system according to claim 8, characterized in that, The steps for analyzing the leaf skeleton morphology heterogeneity index in each region are as follows: The remote sensing image of crops is divided into several regions. The crop leaf skeleton of each region is binarized to obtain a binarized leaf skeleton image. The bifurcation points of the leaf skeleton are identified, and the extension scale, bifurcation deviation angle, and branch topography of each bifurcation point are obtained and summed to obtain the morphological feature values ​​of the two sides of the branches. The shape feature offset values ​​of each leaf skeleton bifurcation point are calculated and compared with the set shape feature offset threshold ρ1 of the bifurcation point. If the shape feature offset of the leaf skeleton bifurcation point is lower than the threshold value ρ1, the bifurcation point is considered to be in the correct position. When the bias value is greater than the morphological bias threshold ρ1, the leaf skeleton bifurcation point is a significant morphological difference bifurcation point. When the bias value of the morphological bias at the leaf skeleton bifurcation point is equal to the morphological bias threshold ρ1, the leaf skeleton bifurcation point is a critical morphological bifurcation point. When the bias value of the morphological bias at the leaf skeleton bifurcation point is less than the morphological bias threshold ρ1, the leaf skeleton bifurcation point is a normal morphological bifurcation point. The number of significant morphological difference bifurcation points, critical morphological bifurcation points, and normal morphological bifurcation points in each region are counted, and then the leaf skeleton morphological heterogeneity index of each region is calculated.

Images