Low-orbit image artifact rapid elimination method based on pulse coupling neural network

Abstract

The invention discloses a low-orbit image artifact rapid elimination method based on a pulse coupling neural network. The method comprises the following steps: S1, acquiring original low-orbit image data; s2, inputting the original low-orbit image data into a satellite-borne hierarchical progressive pulse coupling neural network model; s3, forming an artifact candidate region, and generating an artifact probability graph based on the synchronization pulse connectivity and the information entropy increment; s4, generating preprocessed low-orbit image data, and synchronously generating a reversible repair log; s5, generating an artifact-rejected high-quality image product; and S6, updating the coupling gain coefficient and the link strength threshold of the spaceborne hierarchical progressive pulse coupling neural network model according to the processing state of the artifact-rejected high-quality image product. The system is stable in closed-loop operation, the image quality is remarkably improved, the processing delay meets the quick remote sensing service requirement, and the engineering feasibility is achieved.

Description

Technical Field

[0001] The present invention relates to the field of image processing technology, and in particular, to a method for quickly removing low-orbit image artifacts based on a pulse-coupled neural network. Background Art

[0002] With the rapid growth of the deployment of remote sensing satellite constellations and the increasing demand for high revisit cycle imaging, low-orbit remote sensing satellites have become the main platform for obtaining high-resolution surface observation images. However, due to their relatively low operating orbit altitude (generally between 300 - 1000 km), they are vulnerable to multi-source interference in high-frequency imaging and high-dynamic scenarios, often generating various forms of image artifacts in the imaging data, such as stripe interference, high-energy particle breakdown spots, row and column misalignment, compression error codes, and image jitter caused by attitude perturbations. These artifacts not only seriously interfere with the recognition and analysis of ground object information but also cause systematic errors in subsequent image interpretation, classification, and inversion applications.

[0003] Currently, the mainstream methods for removing image artifacts are mainly divided into two categories: real-time removal at the on-board end and post-processing at the ground end. Most on-board removal methods use fixed threshold filtering, directional template filtering, or median sliding window processing means. Due to limited computing resources, it is difficult to deploy complex deep networks or image segmentation models, resulting in easy failure when facing high-dynamic contrast scenarios or complex texture regions. Although the deep learning artifact repair methods at the ground end have fine effects, due to the fragmentation of the space-ground model structure and the complexity of the recompression processing chain, there are easily problems such as excessive smoothing of image details, structural distortion, or inconsistent textures, and the processing delay is relatively high, making it difficult to meet the requirements for ensuring the quality of low-latency data in disaster monitoring and emergency response scenarios.

[0004] In addition, traditional methods generally lack an end-to-end artifact removal cooperation mechanism, that is, the processing results on the on-board side cannot effectively guide the fine-grained reconstruction at the ground end, and the ground end cannot feedback and adjust the on-board processing parameters, resulting in difficulty in achieving closed-loop optimization of the overall removal strategy. Especially in the case where the artifact distribution is sparse and irregular, scale cross-layer, and texture complex and overlapping, the existing processing frameworks are difficult to balance robustness, real-time performance, and interpretability.

[0005] Therefore, there is an urgent need for a method for quickly removing image artifacts for low-orbit remote sensing satellites, which has real-time performance, reversibility, and space-ground closed-loop adaptive ability, so as to improve the consistency of image quality and the intelligent level of the processing link. Summary of the Invention

[0006] An object of the present invention is to propose a method for quickly removing low-orbit image artifacts based on a pulse-coupled neural network. The system of the present invention operates stably in a closed loop, the image quality is significantly improved, the processing delay meets the requirements of fast remote sensing services, and it has engineering feasibility.

[0007] A method for quickly removing low-orbit image artifacts based on a pulse-coupled neural network according to an embodiment of the present invention includes the following steps: S1. Obtain the original low-orbit image data; S2. Input the original low-orbit image data into the on-board hierarchical progressive pulse-coupled neural network model, and execute pulse traversal according to the synchronization signal of the original low-orbit image to generate a neuron excitation time series; S3. Construct a pulse fire mapping matrix according to the neuron excitation time series, calculate the synchronization pulse connectivity using the pulse fire mapping matrix, form an artifact candidate region, and generate an artifact probability map based on the synchronization pulse connectivity and the information entropy increment; S4. Construct a block-level weight mask according to the artifact probability map, perform pixel repair on the artifact candidate region using the block-level weight mask, generate preprocessed low-orbit image data, and synchronously generate a reversible repair log; S5. Transmit the preprocessed low-orbit image data, the pulse fire mapping matrix, the block-level weight mask, and the reversible repair log to the ground enhancement unit together. The ground enhancement unit performs reversible expansion on the artifact candidate region based on the pulse fire mapping matrix and the reversible repair log, and performs fine-grained texture reconstruction in combination with the block-level weight mask to generate a high-quality image product with removed artifacts; S6. Update the coupling gain coefficient and the link strength threshold of the on-board hierarchical progressive pulse-coupled neural network model according to the processing status of the high-quality image product with removed artifacts, and send the updated parameters back to the on-board hierarchical progressive pulse-coupled neural network model.

[0008] Optionally, S1 includes the following steps: S11. Construct an original low-orbit image data sequence. Each frame of the image represents a visible light image or a multispectral image obtained by a low-orbit remote sensing satellite at a certain moment. The original low-orbit image data sequence consists of image frames obtained at multiple different time points, and each frame of the image is marked with the corresponding acquisition time information; S12. Obtain the attitude quaternion and orbital position of the low-orbit remote sensing satellite corresponding to the acquisition time of each frame of the image. The attitude quaternion and the orbital position together constitute the data basis required for attitude synchronization calibration. Use the attitude quaternion and the orbital position to perform attitude synchronization calibration on the original low-orbit image data sequence. Attitude synchronization calibration is used to register the image frame with the satellite attitude information; S13. Perform preliminary radiometric calibration on the image frames after attitude synchronization calibration. Preliminary radiometric calibration is used to correct the brightness response deviation of the original pixel gray values; S14. After completing the preliminary radiometric calibration, perform strip equalization preprocessing on the image frames to remove large-scale directional brightness non-uniformity features and obtain the final original low-orbit image data.

[0009] Optionally, S2 includes the following steps: S21. Input the final original low-earth orbit image data into the on-board hierarchical progressive pulse coupled neural network model. The on-board hierarchical progressive pulse coupled neural network model includes a global coarse-scale PCNN layer, a medium-scale PCNN layer, and a local fine-scale attention PCNN layer. The three PCNN layers are cascaded hierarchically, and based on the root mean square value of the attitude perturbation of the low-earth orbit remote sensing satellite and the instantaneous gain of the imaging system Initialize the parameters of the global coarse-scale layer: ; Wherein, , , are on-board reference parameters, , , are proportional adjustment factors, is the change amplitude of the attitude perturbation within the rolling observation window, is the global initial coupling gain coefficient, is the global initial adaptive time decay factor, is the global initial link strength threshold, is the reference standard value of the attitude perturbation of the low-earth orbit remote sensing satellite, is the nominal gain value of the imaging system; S22. In the global coarse-scale PCNN layer, drive the update of the internal state of the neurons with the adaptive time decay factor , and adaptively determine the dynamic triggering condition of the neurons according to the global gray-scale statistical variance of the final original low-earth orbit image data, and generate the global coarse-scale neuron pulse output ; S23. In the medium-scale PCNN layer, initialize the medium-scale coupling gain coefficient and the medium-scale initial link strength threshold , and based on the pulse output of the global coarse-scale PCNN layer, calculate the change in local entropy value to obtain the medium-scale dynamic coupling gain coefficient , and dynamically adjust the triggering sensitivity of the medium-scale neurons according to the medium-scale dynamic coupling gain coefficient to form the medium-scale pulse output ; S24. In the local fine-scale attention PCNN layer, initialize the local-scale initial link strength threshold , calculate the local attention coefficient according to the pulse output of the medium-scale PCNN layer, and use the local attention coefficient to dynamically trigger the threshold of the local-scale neuronsPerform real-time adaptive updates to form local fine-scale pulse outputs, and achieve the localization of complex fine-scale artifacts such as stripes, flickering points, compression errors, and particle interferences in low-orbit images; S25. Establish an inter-layer attention interaction mechanism among the global coarse-scale PCNN layer, the medium-scale PCNN layer, and the local fine-scale attention PCNN layer. The pulse output of the global coarse-scale PCNN layer guides the parameter update of the medium-scale PCNN layer, and the pulse output of the medium-scale PCNN layer further guides the parameter update of the local fine-scale attention PCNN layer. The response results of the local fine-scale attention PCNN layer act on the medium-scale and global coarse-scale PCNN layers in reverse, dynamically optimizing the trigger sensitivity and coupling strength of each layer, and achieving a two-way interactive closed loop of coarse-scale-medium-scale-fine-scale and fine-scale-medium-scale-coarse-scale. Finally, a hierarchical progressive neuron excitation time series matrix is generated.

[0010] Optionally, the S3 includes the following steps: S31. Construct a pulse fire mapping matrix based on the hierarchical progressive neuron excitation time series matrix , and the pulse fire mapping matrix is obtained by mapping the excitation time of each pixel point from the hierarchical progressive neuron excitation time series matrix to a two-dimensional matrix consistent with the image coordinates; S32. Based on the pulse fire mapping matrix, calculate the neighborhood pulse synchrony at each pixel position within the receptive field. The neighborhood pulse synchrony is obtained by statistically calculating the relative amplitude of the difference in the first pulse trigger moments between this pixel and all pixels in the neighborhood. The neighborhood pulse synchrony at all positions constitutes a synchronous pulse connectivity graph , and the synchronous pulse connectivity graph is used to characterize the spatial consistency distribution of the neuron excitation timing structure of the original low-orbit image data: ; Among them, is the neighborhood centered on the pixel position , is the total number of pixels in the neighborhood, is the first pulse trigger moment of the neighborhood pixel in the hierarchical progressive attention adaptive pulse coupled neural network model, is the first pulse trigger moment of the neighborhood pixel position in the hierarchical progressive attention adaptive pulse coupled neural network model, is the synchronization degree penalty function, reflecting the degree of inhibition of connectivity by the excitation time difference, represents the strength of the consistency between the pixel position and the neighborhood pulse timing; S33. According to the synchronous pulse connectivity graph The neighborhood pulse synchronization at each pixel position is used to screen out all pixel positions where the neighborhood pulse synchronization is lower than the synchronization demarcation threshold as the preliminary artifact candidate region . The preliminary artifact candidate region is used to identify potential abnormal regions in the original low-orbit image data where the temporal structure of neuron excitation significantly deviates from the surrounding structure. The synchronization demarcation threshold is a fixed empirical threshold for demarcating normal and abnormal synchronization; S34. Calculate the local information entropy increment for each pixel position in the preliminary artifact candidate region . The local information entropy increment is obtained by calculating the difference between the local Shannon entropy within the neighborhood of the pixel position and the average local Shannon entropy of its corresponding synchronous connected region; S35. Combine the neighborhood pulse synchronization and the local information entropy increment of each pixel position to generate an artifact probability map . The artifact probability is obtained by linearly weighting the neighborhood pulse synchronization and the local information entropy increment, where the neighborhood pulse synchronization is used to describe the abnormal temporal structure, and the local information entropy increment is used to describe the abnormal gray structure.

[0011] Optionally, the S4 includes the following steps: S41. Divide the original low-orbit image data into multiple image blocks of a fixed size according to the artifact probability map, and construct a block-level weight mask , where represents the image block coordinates, and the pixel set included in each image block corresponds to the block-level weight mask ; S42. Based on the block-level weight mask , set an artifact repair intensity threshold, and demarcate all image blocks with weight values greater than the artifact repair intensity threshold as the set of repair target block regions ; S43. For each image block belonging to the set of repair target block regions , establish a pixel-level one-to-one correspondence index based on the pulse fire mapping matrix, and perform a pixel-level reversible repair operation to obtain the repaired pixel value . The repaired pixel value is obtained by fusing the original pixel and the neighborhood reference value weighted by the artifact probability; S44. While performing pixel repair, generate a reversible repair log, recording the position index, original pixel value, repaired pixel value, and its corresponding image block index of all repaired pixels; S45. Replace the pixel values at the corresponding positions in the original low-orbit image data with all the repaired pixel values to generate preprocessed low-orbit image data , whose structure remains consistent with the input image frame.

[0012] Optionally, S5 includes the following steps: S51. Transmit the preprocessed low-earth orbit image data, pulse fire mapping matrix, block-level weight mask, and reversible repair log to the ground enhancement unit together. The ground enhancement unit receives the corresponding processing information packet for each frame of the image and constructs a mapping cache structure with a consistent structure; S52. The ground enhancement unit performs pixel-level reversible unfolding operations on the artifact candidate regions according to the reversible repair log and the pulse fire mapping matrix. If a certain pixel position exists in the reversible repair log, take its corresponding original pixel value as the current pixel value. If the pixel position does not exist in the reversible repair log, retain its repaired pixel value as the current pixel value, and finally form a reversible unfolding image data frame; S53. The ground enhancement unit performs fine-grained texture reconstruction operations based on the reversible unfolding image data frame and the weight mask of the corresponding image block to generate new reconstructed output values; S54. Combine the reconstructed output values of all pixel positions into a complete image frame to form an artifact removal high-quality image product output by the ground enhancement unit.

[0013] Optionally, S6 includes the following steps: S61. The ground enhancement unit analyzes the processing status of the artifact removal high-quality image product and extracts the artifact residue index , texture consistency index and gray-scale offset index in the image frame; S62. Normalize the artifact residue index, texture consistency index, and gray-scale offset index and fuse them into a comprehensive artifact removal performance evaluation function ; S63. Construct a space-ground parameter update function based on the change trend of the comprehensive artifact removal performance evaluation function, and dynamically adjust the global coupling gain coefficient and link strength threshold of the spaceborne hierarchical progressive pulse coupled neural network model: ; Among them, , are the coupling gain and threshold update rates respectively, is the expected removal performance target value; S64. Package the updated parameter pair into a parameter download packet, and send it back to the spaceborne processing unit in real time through the space-ground link, and replace the original model parameters for the next round of image data processing to form a closed-loop adaptive iterative update mechanism.

[0014] Optionally, the artifact residue index is: ; Among them, and are the height and width of the image frame, is the pixel position of the artifact probability, represents the average value of the residual artifact probability of the whole image; The texture consistency index: ; Among them, and are respectively the average texture gradients of the th region in the preprocessed and enhanced image frames, is the number of partitions; The gray-scale offset index: ; Among them, is the reconstructed pixel value of the pixel position in the artifact-removed image, is the pixel value in the original image data.

[0015] The beneficial effects of the present invention are: (1) The present invention constructs a hierarchical progressive attention adaptive pulse coupled neural network model with the ability of on-orbit deployment. Without relying on a deep convolutional network, it realizes millisecond-level positioning of sparse and irregular artifacts in low-orbit images. By coupling the global coarse-scale, medium-scale, and local fine-scale pulse propagation layers using a hierarchical PCNN structure and performing dynamic gating through an attention mechanism, each layer can be synchronously regulated according to the local excitation characteristics of the image. By dynamically adjusting the coupling gain coefficient, local threshold, and temporal decay factor, the sensitivity of the model to unstructured artifacts such as particle spots and pseudo-row / column misalignment is significantly improved, ensuring that artifacts can be accurately captured and modeled within 50 ms at an image resolution of 1024×1024.

[0016] (2) The present invention proposes a block-level reversible repair strategy based on pulse fire mapping, realizing high coupling, high fidelity, and reversible mapping between artifact positioning and pixel repair. By constructing a pixel-level indexing system for the pulse fire mapping matrix and the artifact probability map and performing repair credibility regulation based on the image block-level weight mask, a local adaptive fusion function is used to dynamically reconstruct the pixel values of artifacts, and at the same time, a reversible repair log is generated to ensure that the repair status of each pixel position can be fully traced back. While realizing high-fidelity image restoration, it ensures that the ground enhancement algorithm can be carried out as needed and the original texture structure can be reconstructed. Compared with traditional compression coding repair methods, it has stronger interpretability and system compatibility.

[0017] (3) The present invention establishes a satellite-ground bidirectional collaborative update mechanism, dynamically adjusts the core parameters of the pulse-coupled neural network by processing performance feedback, realizes the closed-loop optimization of the artifact removal ability and the image quality control ability, constructs a performance evaluation function that fuses three factors, namely the artifact residue index, the texture consistency index, and the gray-scale offset index, in the ground enhancement unit, and updates the global coupling gain coefficient and the link strength threshold in the spaceborne model in real time through the evaluation result, establishes a satellite-ground feedback closed-loop, can adapt to the image quality changes of the satellite under different orbital segments and different imaging conditions, realizes the dynamic balance between artifact suppression and the retention of the original texture, and improves the image processing consistency under attitude disturbances. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The drawings are used to provide a further understanding of the present invention, and constitute a part of the specification, and are used to explain the present invention together with the embodiments of the present invention, and do not constitute a limitation to the present invention. In the drawings: Figure 1 is a flowchart of a method for quickly removing low-orbit image artifacts based on a pulse-coupled neural network proposed by the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0019] Now, the present invention will be further described in detail with reference to the drawings. These drawings are all simplified schematic diagrams, and only illustrate the basic structure of the present invention in a schematic manner, so they only show the components related to the present invention.

[0020] Refer to Figure 1 , a method for quickly removing low-orbit image artifacts based on a pulse-coupled neural network, includes the following steps: S1. Obtain the original low-orbit image data, which is generated by the low-orbit remote sensing satellite after attitude synchronous calibration and preliminary radiometric calibration are completed; S2. Input the original low-orbit image data into the spaceborne hierarchical progressive pulse-coupled neural network model, initialize the coupling gain coefficient, the link strength threshold, and the time decay parameter, and generate the neuron firing time series according to the synchronous signal execution pulse traversal based on the original low-orbit image; S3. Construct a pulse fire mapping matrix according to the neuron firing time series. The pulse fire mapping matrix records the pulse trigger order and the neighborhood connectivity relationship at each pixel position, calculates the synchronous pulse connectivity using the pulse fire mapping matrix, forms an artifact candidate region, and generates an artifact probability map based on the synchronous pulse connectivity and the information entropy increment; S4. Construct a block-level weight mask according to the artifact probability map. The block-level weight mask and the pulse fire mapping matrix establish a one-to-one correspondence index to achieve repair reversibility. Use the block-level weight mask to perform pixel repair on the artifact candidate region, generate preprocessed low-orbit image data, and synchronously generate a reversible repair log. The reversible repair log records the pixel values before and after repair and the corresponding indexes; S5. Transmit the pre - processed low - earth - orbit image data, the pulse fire mapping matrix, the block - level weight mask, and the reversible repair log to the ground enhancement unit together. The ground enhancement unit performs reversible unfolding on the artifact candidate regions based on the pulse fire mapping matrix and the reversible repair log, and performs fine - grained texture reconstruction in combination with the block - level weight mask to generate an artifact - removed high - quality image product; S6. Update the coupling gain coefficient and the link strength threshold of the on - satellite hierarchical progressive pulse - coupled neural network model according to the processing status of the artifact - removed high - quality image product, and send the updated parameters back to the on - satellite hierarchical progressive pulse - coupled neural network model to form a space - ground closed - loop adaptive artifact - removal process.

[0021] In this embodiment, S1 includes the following steps: S11. Construct an original low - earth - orbit image data sequence. Each frame of the image represents a visible - light image or a multi - spectral image acquired by the low - earth - orbit remote - sensing satellite at a certain moment. The original low - earth - orbit image data sequence consists of image frames acquired at multiple different time points, and each frame of the image is marked with the corresponding acquisition time information; S12. Obtain the attitude quaternion and the orbital position of the low - earth - orbit remote - sensing satellite corresponding to the acquisition time of each frame of the image. The attitude quaternion represents the attitude information of the low - earth - orbit remote - sensing satellite in three - dimensional space, and the orbital position represents the position coordinates of the low - earth - orbit remote - sensing satellite in inertial space. The attitude quaternion and the orbital position together constitute the data basis required for attitude synchronization calibration. Use the attitude quaternion and the orbital position to perform attitude synchronization calibration on the original low - earth - orbit image data sequence. Attitude synchronization calibration is used to register the image frame with the satellite attitude information, so that each pixel point in the image establishes a spatially consistent relationship with the physical ground position; S13. Perform preliminary radiometric calibration on the image frames after attitude synchronization calibration. Preliminary radiometric calibration is used to correct the brightness response deviation of the original pixel gray values. The brightness response value is obtained by multiplying the radiation gain coefficient and adding the radiation offset term. The radiation gain coefficient is used to characterize the amplification ratio of the imaging system to the input signal, and the radiation offset term is used to correct the zero - point drift of the sensor. The brightness response value serves as the pixel intensity basis of the image in the physical radiation sense; S14. After completing the preliminary radiometric calibration, perform strip - equalization pre - processing on the image frames. The strip - equalization pre - processing is carried out along the column direction of the image. Calculate the average value of all pixel values in each column of the image frame to obtain the column average value. The column average value represents the overall brightness level of the column in the vertical direction. The column average value is used to judge the offset trend of the directional brightness non - uniformity, and the column average value is used for pixel intensity normalization processing to remove the large - scale directional brightness non - uniformity characteristics and obtain the final original low - earth - orbit image data.

[0022] In this embodiment, S2 includes the following steps: S21. Input the final original low-earth orbit image data into the on-board hierarchical progressive pulse-coupled neural network model. The on-board hierarchical progressive pulse-coupled neural network model includes a global coarse-scale PCNN layer, a medium-scale PCNN layer, and a local fine-scale attention PCNN layer. The three PCNN layers are cascaded, and based on the root mean square value of the attitude perturbation of the low-earth orbit remote sensing satellite and the instantaneous gain of the imaging system Initialize the parameters of the global coarse-scale layer: ; Among them, , , are on-board reference parameters, , , are proportional adjustment factors, is the change amplitude of the attitude perturbation within the rolling observation window, is the global initial coupling gain coefficient, is the global initial adaptive time decay factor, is the global initial link strength threshold, is the reference standard value of the attitude perturbation of the low-earth orbit remote sensing satellite, is the nominal gain value of the imaging system; S22. In the global coarse-scale PCNN layer, drive the update of the internal state of the neurons with the adaptive time decay factor , and adaptively determine the dynamic triggering conditions of the neurons according to the global gray-scale statistical variance of the final original low-earth orbit image data, and generate the global coarse-scale neuron pulse output for the preliminary positioning of large-scale strips and obvious artifacts caused by high-energy particle breakdown in the original low-earth orbit image data: ; Among them, is the internal state voltage of the pixel position at time , is the pixel gray value of the final original low-earth orbit image data, is the receptive field neighborhood, is the long-range coupling weight, is the pulse output of the previous moment, is the global variance amplification coefficient; S23. In the medium-scale PCNN layer, initialize the medium-scale coupling gain coefficient and the medium-scale initial link strength threshold , and calculate the local entropy value change based on the pulse output of the global coarse-scale PCNN layer to obtain the medium-scale dynamic coupling gain coefficient , dynamically adjust the triggering sensitivity of mesoscale neurons according to the mesoscale dynamic coupling gain coefficient to form a mesoscale pulse output for capturing the edge and contour features of low-orbit image artifacts: ; The mesoscale dynamic coupling gain coefficient is calculated as: ; where is the entropy sensitivity factor, is the entropy upper limit, is the pixel position neighborhood gray level instantaneous probability of is the set of gray levels, is the coarse-layer pulse correction weight; S24. In the local fine-scale attention PCNN layer, initialize the local scale initial link strength threshold , calculate the local attention coefficient according to the pulse output of the mesoscale PCNN layer, and use the local attention coefficient to perform real-time adaptive update on the dynamic trigger threshold of the local scale neurons to form a local fine-scale pulse output for locating complex fine-scale artifacts such as stripes, flickering points, compression errors, and particle interference in low-orbit images: ; The dynamic trigger threshold is calculated as: ; where is neighborhood, is the number of neighborhood pixels, is the mesoscale pulse output, is the local reference threshold, is the attention sensitivity factor, is the gradient gain coefficient, is the pixel gray gradient magnitude, is the coupling feedback weight; S25. An inter-layer attention interaction mechanism is established among the global coarse-scale PCNN layer, the medium-scale PCNN layer, and the local fine-scale attention PCNN layer. The pulse output of the global coarse-scale PCNN layer guides the parameter update of the medium-scale PCNN layer, and the pulse output of the medium-scale PCNN layer further guides the parameter update of the local fine-scale attention PCNN layer. The response result of the local fine-scale attention PCNN layer acts on the medium-scale and global coarse-scale PCNN layers in reverse, dynamically optimizing the trigger sensitivity and coupling strength of each layer, realizing a two-way interaction closed-loop of coarse-scale-medium-scale-fine-scale and fine-scale-medium-scale-coarse-scale, and finally generating a hierarchical progressive neuron firing time series matrix.

[0023] In this embodiment, S3 includes the following steps: S31. Construct a pulse fire mapping matrix according to the hierarchical progressive neuron firing time series matrix . The pulse fire mapping matrix records the first pulse trigger moment of the neuron corresponding to each pixel position in the original low-orbit image data. The first pulse trigger moment is used to measure the response priority of the neuron at this position and its positioning order in the time sequence structure. The pulse fire mapping matrix is obtained by mapping the firing time of each pixel point from the hierarchical progressive neuron firing time series matrix to a two-dimensional matrix consistent with the image coordinates; S32. Based on the pulse fire mapping matrix, calculate the neighborhood pulse synchrony within the receptive field of each pixel position. The neighborhood pulse synchrony is used to characterize the degree of consistency in the pulse trigger time sequence between this pixel position and its neighboring pixels. The neighborhood pulse synchrony is obtained by statistically calculating the relative amplitude of the difference in the first pulse trigger moments between this pixel and all pixels in the neighborhood. The neighborhood pulse synchrony of all positions constitutes a synchronous pulse connectivity graph , and the synchronous pulse connectivity graph is used to represent the spatial consistency distribution of the neuron firing time sequence structure of the original low-orbit image data: ; Among them, is the neighborhood centered on the pixel position , is the total number of pixels in the neighborhood, is the first pulse trigger moment of the neighborhood pixel in the hierarchical progressive attention adaptive pulse-coupled neural network model, is the first pulse trigger moment of the neighborhood pixel position in the hierarchical progressive attention adaptive pulse-coupled neural network model, is the synchronization degree penalty function, reflecting the degree of inhibition of connectivity by the excitation time difference, represents the strength of the consistency between the pixel position and the neighborhood pulse time sequence; S33. According to the synchronous pulse connection diagram The neighborhood pulse synchronization of each pixel position in the image is filtered out, and all the neighborhood pulse synchronizations below the synchronization threshold are screened out. The pixel position of The preliminary candidate region of the artifact is used to identify the potential abnormal region where the timing structure of neuronal firing is obviously separated from the surrounding structure in the original low-orbit image data. The synchronization demarcation threshold A fixed empirical threshold for demarcating normal and abnormal synchrony; S34. Calculate the local information entropy increment for each pixel position in the preliminary candidate region of the artifact The local information entropy increment is used to measure the deviation of the grayscale complexity of the local area where the pixel is located from the average grayscale complexity of the synchronously connected area to which it belongs. The local information entropy increment is obtained by calculating the difference between the local Shannon entropy in the neighborhood of the pixel position and the average local Shannon entropy of its corresponding synchronously connected area. The larger the local information entropy increment, the more inconsistent the grayscale structure of the pixel position is, and the more likely it is that there is image artifact interference; S35. Generate an artifact probability map by combining the neighborhood pulse synchronization and local information entropy increment of each pixel position , the artifact probability map is used to indicate the probability of an artifact at each pixel position. The artifact probability is obtained by linearly weighting the neighborhood pulse synchronization and the local information entropy increment, where the neighborhood pulse synchronization is used to describe the abnormal timing structure, and the local information entropy increment is used to describe the abnormal grayscale structure: ; in, , is the linear weighting factor, is the normalized upper bound of the local entropy set.

[0024] In this implementation, S4 includes the following steps: S41. Divide the original LEO image data into multiple fixed-size image blocks according to the artifact probability map, and construct a block-level weight mask ,in Represents the image block coordinates, each image block contains a set of pixels The corresponding block-level weight mask is defined as: ; in, Represents the total number of pixels in the image block, is the pixel position Artifact probability value, block-level weight mask Indicates the overall artifact intensity of the image block; S42. On the block-level weight mask Based on this, set the artifact repair intensity threshold, and delimit all image blocks with weight values greater than the artifact repair intensity threshold as the set of repair target block regions ; S43. For each image block belonging to the set of repair target block regions Based on the impulse fire mapping matrix, establish a pixel-level one-to-one correspondence index, and perform a pixel-level reversible repair operation to obtain the repaired pixel value , and the repaired pixel value is obtained by weighted fusion of the original pixel and the neighborhood reference value according to the artifact probability: ; Among them, represents the artifact weight of the current pixel, is the original pixel value, represents the pixel position the mean value of the neighborhood pixels; S44. While performing pixel repair, generate a reversible repair log, recording the position index, original pixel value, repaired pixel value of all repaired pixels, and the index of the image block where they are located; S45. Replace the pixel values at the corresponding positions in the original low-orbit image data with all the repaired pixel values to generate preprocessed low-orbit image data , and its structure remains consistent with the input image frame.

[0025] In this embodiment, S5 includes the following steps: S51. Transmit the preprocessed low-orbit image data, the impulse fire mapping matrix, the block-level weight mask, and the reversible repair log to the ground enhancement unit together. The ground enhancement unit receives the corresponding processing information packet of each frame of image and constructs a mapping cache structure with a consistent structure; S52. The ground enhancement unit performs a pixel-level reversible expansion operation on the artifact candidate region according to the reversible repair log and the impulse fire mapping matrix. The pixel-level reversible expansion operation is used to restore each repaired pixel value. The restoration is based on the original pixel value before repair recorded in the reversible repair log. If the original pixel value corresponding to a certain pixel position exists in the reversible repair log, take it as the current pixel value. If the pixel position does not exist in the reversible repair log, keep its repaired pixel value as the current pixel value, and finally form a reversibly expanded image data frame; S53. The ground enhancement unit performs a fine-grained texture reconstruction operation based on the reversibly expanded image data frame and the weight mask of the corresponding image block. The fine-grained texture reconstruction operation is used to fuse the texture credibility weight of the image block where the pixel is located with the weighted average of the cross-frame observed pixels to generate a new reconstructed output value. The texture credibility weight is obtained by subtracting the block-level weight mask value of the image block to which the pixel belongs from 1, indicating the degree of the pixel area being a credible texture. The weighted average of the cross-frame observed pixels is used to introduce multi-frame redundant information to enhance the image texture quality. The generated reconstructed output value is the weighted fusion result of the current pixel value and the cross-frame average value; S54. Combine the reconstructed output values at all pixel positions into a complete image frame to form an artifact removal high-quality image product output by the ground enhancement unit.

[0026] In this embodiment, S6 includes the following steps: S61. The ground enhancement unit analyzes the processing state of the artifact removal high-quality image product and extracts the artifact residue index in the image frame , the texture consistency index and the gray-scale offset index ; S62. Normalize and fuse the artifact residue index, the texture consistency index, and the gray-scale offset index into a comprehensive artifact removal performance evaluation function : ; Among them, , , are respectively the index normalization values, , , is the weighting coefficient; S63. Construct a satellite-ground parameter update function based on the change trend of the comprehensive artifact removal performance evaluation function, and dynamically adjust the global coupling gain coefficient and the link strength threshold of the spaceborne hierarchical progressive pulse coupled neural network model: ; Among them, , are respectively the coupling gain and the threshold update rate, is the expected artifact removal performance target value; S64. Package the updated parameter pair into a parameter download packet, and send it back to the spaceborne processing unit in real time through the satellite-ground link, and replace the original model parameters for the next round of image data processing, forming a closed-loop adaptive iterative update mechanism to realize the dynamic optimization of the satellite-ground integrated artifact removal performance.

[0027] In this embodiment, the artifact residue index is as follows: ; where and are the height and width of the image frame, is the pixel position of the artifact probability, represents the average value of the residual artifact probability of the whole image; Texture consistency index: ; where and are respectively the average texture gradients of the th region in the preprocessed and enhanced image frames, is the number of partitions; Gray-scale offset index: ; where is the reconstructed pixel value at position in the artifact-removed image, is the pixel value in the original image data.

[0028] Example 1: In the continuous downlink image data stream, when a low-earth orbit remote sensing satellite processes five frames of image data numbered 010438–010442, the system monitors that there are significant strip artifacts in the column direction of these images. The artifact area is mainly concentrated in the lower left quadrant of the image. The strip width is about 6–12 pixels, and the average brightness fluctuation amplitude exceeds 1.8 times that of the surrounding background. The initial detection of the artifact is triggered by the on-board synchronization signal trigger mechanism. After each frame of image undergoes pulse traversal, multiple consecutive synchronous "pulse-dense" strip blocks appear in the neuron firing time series, and the trigger times are highly close. The system automatically marks these areas as high-risk trigger areas.

[0029] Subsequently, the system constructs a pulse fire mapping matrix based on this firing time series and calculates the synchronous pulse connectivity under the neighborhood. For the 010440th frame of image, the system identifies a low-synchronization area with a connectivity less than 0.42 within the pixel coordinate range (314:320, 240:410), and the area exceeds 2.9% of the total pixels of the image. At the same time, the average value of the local entropy increment in this area reaches 0.161, which is 3.4 times that of other areas of the image. The system constructs an artifact probability map based on this, and the maximum probability point appears at position (317, 252), and the artifact probability is 0.91.

[0030] According to the artifact probability map, the image is divided into Image blocks are obtained, a total of 256 image blocks. Among them, 19 image blocks have block-level weight mask values exceeding 0.75 and are designated as severely artifact-contaminated areas by the system. These image blocks are then submitted to the pixel repair engine for processing. A log record is generated for each repaired pixel point, including the value before repair, the value after repair, and the corresponding image frame, image block, and coordinate information. For example, in the image block numbered B-118, the gray value at pixel position (321, 260) is repaired from the original 136 to 122, with a weight coefficient of 0.88, and is recorded in the reversible repair log entry R-0047591.

[0031] All repair logs and fire mapping matrices are downloaded to the ground processing module together with the preprocessed image. After the ground module receives image 010440, by looking up the repair logs and weight masks in reverse, it is found that the regional repair coefficients of image blocks B-118, B-120, and B-135 are relatively high and the artifact edges with inconsistent frequencies still appear after texture reconstruction. The system starts the fine-grained texture enhancement process. First, the above-mentioned image blocks are reversibly unfolded to restore the original pixels, and then multiple frames of images in the corresponding regions in the past 3 track segments are called. Through the method of sliding weighted averaging in a time window, the temporal smoothing value of each pixel is fused. For example, the weighted average value of pixel (321, 260) is 127.4, and the final fused output value is 125.

[0032] After the ground enhancement output is completed, the system performs an assessment of the artifact residue and texture consistency for this frame of image. The calculated artifact residue index is 0.036, a 58% decrease compared to the previous version; the texture consistency index is increased from 0.86 to 0.93; the average repair gray offset is reduced from 11.2 to 4.3; the relative mean error of the image is less than 2.4%. At the same time, the comprehensive artifact removal performance function of this image frame obtains an updated value of 0.128, which is lower than the system-set threshold of 0.15.

[0033] According to the performance scoring results, the ground system automatically adjusts the coupling gain coefficient and synchronization threshold in the on-board model. The new parameter pairs are and , and the system encodes this parameter update package and sends it back to the on-board processing end, numbered UPD-01883, which has taken effect during the processing of image frame numbered 010443.

[0034] The image frame after this parameter update is displayed in the downlink data. The high-value areas in the overall artifact probability map are significantly reduced. The weight mask value of image block B-118 is reduced from the original 0.81 to 0.44. At the same time, image block B-120 is no longer marked as a candidate area after processing.

[0035] After the entire process is completed, the evaluation report output by the system shows that in this example, a total of 107,392 pixel points of suspected stripe artifacts are removed, accounting for approximately 10.2% of the total image area. Finally, the average artifact residue probability drops to 0.039, a 65.8% decrease compared to the control group images without this method (0.114); the texture structure similarity index SSIM increases by 12.7%, and the accuracy of manual review increases by 15.3%.

[0036] Example 1 of the present invention verifies the recognition ability and automatic repair efficiency of the method of the present invention for stripe, bright spot, and dynamic compression artifacts under high-interference conditions. At the same time, through satellite-ground collaborative feedback, adaptive control of the core parameters of the model is achieved. The system operates stably in a closed loop, the image quality is significantly improved, the processing delay meets the requirements of fast remote sensing services, and it has engineering feasibility.

[0037] The present invention constructs a hierarchical progressive attention adaptive pulse-coupled neural network model with the ability to be deployed on satellites. Without relying on a deep convolutional network, it realizes millisecond-level positioning of sparse and irregular artifacts in low-earth orbit images. It uses a hierarchical PCNN structure to couple the global coarse-scale, medium-scale, and local fine-scale pulse propagation layers, and performs dynamic gating through an attention mechanism, enabling each layer to perform synchronous regulation according to the local excitation characteristics of the image. By dynamically adjusting the coupling gain coefficient, local threshold, and temporal decay factor, the sensitivity of the model to unstructured artifacts such as particle spots and pseudo-row and column misalignments is significantly improved, ensuring that artifacts can be accurately captured and modeled within 50 ms at an image resolution of 1024×1024.

[0038] The present invention proposes a block-level reversible repair strategy based on pulse fire mapping, realizing high coupling, high fidelity, and reversible mapping between artifact positioning and pixel repair. By constructing a pixel-level indexing system for the pulse fire mapping matrix and the artifact probability map, and performing repair credibility regulation based on the image block-level weight mask, a local adaptive fusion function is used to dynamically reconstruct the pixel values of the artifacts, while generating a reversible repair log to ensure that the repair status of each pixel position can be completely traced back. While achieving high-fidelity image restoration, it ensures that the ground enhancement algorithm can be deployed as needed and reconstruct the original texture structure, and has stronger interpretability and system compatibility compared to traditional compression coding repair methods.

[0039] The present invention establishes a space-ground bidirectional collaborative update mechanism, dynamically adjusts the core parameters of the pulse-coupled neural network by processing performance feedback, realizes the closed-loop optimization of the artifact rejection ability and the image quality control ability, constructs a performance evaluation function that fuses three factors, namely, the artifact residue index, the texture consistency index, and the gray-scale offset index, in the ground enhancement unit, and updates the global coupling gain coefficient and the link strength threshold in the spaceborne model in real time according to the evaluation result, establishes a space-ground feedback closed loop, can adapt to the image quality changes of the satellite under different orbital segments and different imaging conditions, realizes the dynamic balance between artifact suppression and original texture retention, and improves the image processing consistency under attitude disturbances.

[0040] As described above, only the specific preferred embodiments of the present invention are given, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, according to the technical solution and inventive concept of the present invention, makes equivalent substitutions or changes, and all should be covered within the protection scope of the present invention.

Claims

1. A method for fast removal of low-orbit image artifacts based on pulse coupled neural network, characterized in that: The steps include: S1. Obtaining original low-orbit image data; S2. Input the original LEO image data into the onboard hierarchical progressive pulse coupled neural network model, and perform pulse traversal based on the synchronization signal of the original LEO image to generate a neuron excitation time series; S3. construct a pulse fire mapping matrix according to the neuron excitation time series, calculate the synchronization pulse connectivity using the pulse fire mapping matrix, form an artifact candidate area, and generate an artifact probability map based on the synchronization pulse connectivity and information entropy increment; S4. construct a block-level weight mask according to the artifact probability map, perform pixel restoration on the artifact candidate area using the block-level weight mask, generate pre-processed low-orbit image data, and simultaneously generate a reversible restoration log; S5. The pre-processed low-orbit image data, the pulse fire mapping matrix, the block-level weight mask and the reversible repair log are transmitted to the ground enhancement unit. The ground enhancement unit reversibly expands the candidate artifact area based on the pulse fire mapping matrix and the reversible repair log, and performs fine-grained texture reconstruction in combination with the block-level weight mask to generate a high-quality image product with artifact removal; S6. Update the coupling gain coefficient and link strength threshold of the onboard hierarchical progressive pulse coupled neural network model according to the processing status of the artifact-removed high-quality image product, and return the updated parameters to the onboard hierarchical progressive pulse coupled neural network model.

2. According to the method of claim 1, which is characterized by: The S1 comprises the following steps: S11. Constructing an original low-orbit image data sequence, where each frame of image represents a visible light image or a multispectral image acquired by a low-orbit remote sensing satellite at a certain moment. The original low-orbit image data sequence is composed of image frames acquired at multiple different time points, and each frame of image is marked with corresponding acquisition time information; S12. Acquire the attitude quaternion and orbital position of the low-orbit remote sensing satellite corresponding to each frame image acquisition moment, the attitude quaternion and the orbital position together constitute the data basis required for attitude synchronization calibration, the attitude quaternion and the orbital position are used to perform attitude synchronization calibration on the original low-orbit image data sequence, and the attitude synchronization calibration is used to align the image frame with the satellite attitude information; S13. Performing preliminary radiation calibration on the image frame after attitude synchronization calibration, the preliminary radiation calibration is used to correct the brightness response deviation of the original pixel gray value; S14. After completing the preliminary radiation calibration, perform strip equalization preprocessing on the image frame to remove large-scale directional brightness unevenness features to obtain the final original low-orbit image data.

3. According to the method of claim 1, which is characterized by: The S2 comprises the following steps: S21. Input the final raw low-orbit image data into the onboard hierarchical progressive pulse coupled neural network model. The onboard hierarchical progressive pulse coupled neural network model includes a global coarse-scale PCNN layer, a medium-scale PCNN layer, and a local fine-scale attention PCNN layer. The three layers of PCNN are cascaded and based on the root mean square value of the low-orbit remote sensing satellite attitude disturbance The instantaneous gain of the imaging system Initialize the global coarse-scale layer parameters: ； in, , , is the onboard reference parameter, , , is the proportional adjustment factor, is the amplitude of attitude disturbance change in the rolling observation window, is the global initial coupling gain coefficient, is the global initial adaptive time decay factor, is the global initial link strength threshold, is the reference standard value of attitude disturbance of low-orbit remote sensing satellite, is the nominal gain value of the imaging system; S22. In the global coarse-scale PCNN layer, the adaptive time decay factor Drive the internal state of the neuron to update, and according to the global grayscale statistical variance of the final original low-orbit image data Adaptively determine the dynamic triggering conditions of neurons and generate global coarse-scale neuron pulse outputs ; S23. Initialize the mesoscale coupling gain coefficient in the mesoscale PCNN layer and the mesoscale initial link strength threshold , and based on the pulse output of the global coarse-scale PCNN layer, calculate the local entropy value change Obtaining the mesoscale dynamic coupling gain coefficient , according to the mesoscale dynamic coupling gain coefficient Dynamically adjust the trigger sensitivity of mesoscale neurons to form mesoscale pulse output ; S24. Initialize the local scale initial link strength threshold in the local fine-scale attention PCNN layer , the local attention coefficient is calculated based on the pulse output of the mid-scale PCNN layer , and use the local attention coefficient Dynamic triggering thresholds for local-scale neurons Perform real-time adaptive updates to form local fine-scale pulse outputs, and locate complex fine-scale artifacts such as stripes, scintillation points, compression errors, and particle interference in low-orbit images; S25. An inter-layer attention interaction mechanism is established between the global coarse-scale PCNN layer, the middle-scale PCNN layer and the local fine-scale attention PCNN layer. The pulse output of the global coarse-scale PCNN layer guides the parameter update of the middle-scale PCNN layer, and the pulse output of the middle-scale PCNN layer guides the parameter update of the local fine-scale attention PCNN layer. The response result of the local fine-scale attention PCNN layer acts reversely on the middle-scale and global coarse-scale PCNN layers, dynamically optimizing the trigger sensitivity and coupling strength of each layer, realizing a two-way interactive closed loop of coarse-scale-middle-scale-fine-scale and fine-scale-middle-scale-coarse-scale, and finally generating a hierarchical progressive neuron excitation time series matrix.

4. According to claim 3, a method for fast removal of low-orbit image artifacts based on pulse coupled neural network is characterized in that: The S3 comprises the following steps: S31. Constructing a pulse-fire mapping matrix based on the hierarchical progressive neuron firing time series matrix ,The pulse fire mapping matrix is ​​obtained by mapping the firing time of each pixel from the hierarchical ,neuron firing time series matrix to a two-dimensional matrix consistent with the ,image coordinates; S32. Based on the pulse fire mapping matrix, the neighborhood pulse synchronization of each pixel position within the receptive field is calculated. The neighborhood pulse synchronization is obtained by counting the relative amplitude of the difference between the first pulse triggering time of the pixel and all pixels in the neighborhood. The neighborhood pulse synchronization of all positions constitutes a synchronous pulse connectivity graph. , the synchronous spike connectivity map is used to characterize the spatial consistency distribution of the neuronal firing temporal structure of the original LEO image data: ； in, The pixel position centered Neighborhood, is the total number of pixels in the neighborhood, Neighborhood pixels The first pulse triggering moment in the hierarchical progressive attention adaptive pulse coupled neural network model, is the neighborhood pixel position The first pulse triggering moment in the hierarchical progressive attention adaptive pulse coupled neural network model, is the synchronization penalty function, which reflects the degree of inhibition of connectivity caused by the excitation time difference. Indicates pixel position The strength of the consistency with the neighboring pulse timing; S33. According to the synchronous pulse connection diagram The neighborhood pulse synchronization of each pixel position in the image is filtered out, and all the neighborhood pulse synchronizations below the synchronization threshold are screened out. The pixel position of The preliminary candidate region of the artifact is used to identify the potential abnormal region where the timing structure of neuronal firing is obviously separated from the surrounding structure in the original low-orbit image data. The synchronization demarcation threshold A fixed empirical threshold for demarcating normal and abnormal synchrony; S34. Calculate the local information entropy increment for each pixel position in the preliminary candidate region of the artifact , the local information entropy increment is obtained by calculating the difference between the local Shannon entropy in the neighborhood of the pixel position and the average local Shannon entropy of its corresponding synchronously connected area; S35. Generate an artifact probability map by combining the neighborhood pulse synchronization and local information entropy increment of each pixel position , the artifact probability is obtained by linearly weighting the neighborhood pulse synchronization and the local information entropy increment, where the neighborhood pulse synchronization is used to describe the abnormal timing structure, and the local information entropy increment is used to describe the abnormal grayscale structure.

5. The method for fast removing artifacts from low-orbit images based on pulse coupled neural network according to claim 4 is characterized in that: The S4 comprises the following steps: S41. Divide the original LEO image data into multiple fixed-size image blocks according to the artifact probability map, and construct a block-level weight mask ,in Represents the image block coordinates, each image block contains a set of pixels Corresponding block-level weight mask ; S42. Weight mask at block level On this basis, the artifact restoration strength threshold is set, and all image blocks with weight values ​​greater than the artifact restoration strength threshold are defined as the restoration target block area set. ; S43. For each set of repair target block areas The image block is constructed by establishing a one-to-one corresponding index at the pixel level based on the pulse fire mapping matrix, and a pixel-level reversible repair operation is performed to obtain the repaired pixel value. ,The restored pixel value is obtained by weighted fusion of the original pixel and the neighborhood reference value according to the artifact probability; S44. While performing pixel restoration, generate a reversible restoration log to record the position index, original pixel value, restored pixel value and image block index of all restored pixels; S45. Set all the restored pixel values Replace the pixel values ​​at the corresponding positions in the original LEO image data to generate pre-processed LEO image data , whose structure remains consistent with the input image frame.

6. The method for fast removing artifacts from low-orbit images based on pulse coupled neural network according to claim 5 is characterized in that: The S5 comprises the following steps: S51. The pre-processed low-orbit image data, the pulse fire mapping matrix, the block-level weight mask and the reversible repair log are transmitted to the ground enhancement unit, and the ground enhancement unit receives the corresponding processing information packet of each frame of the image and constructs a mapping cache structure with a consistent structure; S52. The ground enhancement unit performs a pixel-level reversible expansion operation on the artifact candidate area according to the reversible restoration log and the pulse fire mapping matrix. If a certain pixel position exists in the reversible restoration log, the corresponding original pixel value is taken as the current pixel value. If the pixel position does not exist in the reversible restoration log, the restored pixel value is retained as the current pixel value, and finally a reversible expansion image data frame is formed; S53. The ground enhancement unit performs a fine-grained texture reconstruction operation based on the reversibly expanded image data frame and the weight mask of the corresponding image block to generate a new reconstructed output value; S54. The reconstructed output values ​​of all pixel positions are combined into a complete image frame, which constitutes a high-quality image product with artifacts removed output by the ground enhancement unit.

7. The method for fast removing artifacts from low-orbit images based on pulse coupled neural network according to claim 6 is characterized in that: The S6 comprises the following steps: S61. The ground enhancement unit analyzes the processing status of the artifact-removed high-quality image product and extracts the artifact residual index in the image frame , Texture consistency index Grayscale deviation index ; S62. Normalize the artifact residue index, texture consistency index and grayscale offset index and fuse them into a comprehensive artifact removal performance evaluation function ; S63. Based on the changing trend of the comprehensive artifact rejection performance evaluation function, the satellite-ground parameter update function is constructed to dynamically adjust the global coupling gain coefficient of the satellite-borne hierarchical progressive pulse coupled neural network model. Link strength threshold : ； in, , are coupling gain and threshold update rate respectively, is the desired culling performance target value; S64. Update the parameters to The parameters are packaged and sent back to the onboard processing unit in real time through the satellite-to-ground link, and the original model parameters are replaced for the next round of image data processing, forming a closed-loop adaptive iterative update mechanism.

8. The method for fast removing artifacts from low-orbit images based on pulse coupled neural network according to claim 7 is characterized in that: The artifact residual index is: ； in, , is the height and width of the image frame, is the pixel position The probability of artifacts, Represents the mean probability of residual artifacts in the whole image; The texture consistency index: ； in, , Respectively The mean texture gradient of the region in the preprocessed and enhanced image frame, is the number of partitions; The grayscale deviation index: ； in, Pixel positions in the image to remove artifacts The reconstructed pixel value, is the pixel value in the original image data.

Images