Satellite image compression transmission method and system for reservoir monitoring station image transmission

Abstract

The application relates to the technical field of image compression transmission, and discloses a satellite image transmission monitoring station image compression transmission method and system for a reservoir, which comprises the following steps: obtaining an original image, removing system errors to obtain a clear image; extracting a waveband edge intensity and a texture gradient, performing a connected domain morphological close operation after double-threshold screening, and determining a core target region boundary; expanding an adjacent pixel outward by counting a boundary pixel histogram to obtain an expansion region; calculating a saliency map based on the region, screening high saliency pixels, performing connected component marking, and generating a protection area mask; performing wavelet transformation on the expansion region, encoding high-fidelity data according to a mask low quantization step, high quantization encoding of a background region to generate compressed data, and fusing the high-fidelity data and the compressed data to obtain a hybrid compressed image; after verification, when a receiving end decodes, core data is losslessly restored, a background region is reconstructed in detail, and a complete image is output. The method can solve the problem of image distortion.

Description

Technical Field

[0001] This invention relates to the field of image compression and transmission technology, and in particular to an image compression and transmission method and system for a satellite image transmission monitoring station for a reservoir. Background Technology

[0002] Currently, reservoir monitoring relies on the efficient analysis and mining of big data processing technology to accurately extract, integrate, and interpret information related to reservoir status.

[0003] In one existing technology, a unified image compression and transmission scheme based on a fixed bit rate is adopted. The entire reservoir image is acquired and preprocessed. The preprocessed image of the entire reservoir is input into the compression algorithm, and all pixel areas of the image are compressed and encoded without difference. Core detail areas and background areas are compressed with the same intensity. The compressed image data is split into fixed data packets. Based on the fixed transmission protocol of satellite communication, all data packets are transmitted to the ground reservoir monitoring terminal in sequence using an equal resource allocation method.

[0004] However, existing technologies use a uniform compression intensity for the entire image, often sacrificing details of critical areas such as dam cracks and water level fluctuations in pursuit of high compression rates. This can lead to misjudgments in subsequent safety analyses. Furthermore, existing technologies lack targeted processing for the complex environment of reservoirs, making them susceptible to topographic distortion, lighting changes, and noise interference, resulting in image distortion or misjudgments of core areas. Additionally, the lack of a quality verification process before transmission can lead to issues such as excessive compression distortion that still results in transmission, rendering the image unusable at the receiving end. In summary, existing technologies suffer from image distortion problems. Summary of the Invention

[0005] This invention provides a method and system for image compression and transmission at a satellite image transmission monitoring station for reservoirs, in order to solve the problem of image distortion in the prior art.

[0006] Firstly, in order to solve the above-mentioned technical problems, the present invention provides an image compression and transmission method for a satellite image transmission monitoring station for reservoirs, comprising:

[0007] Acquire multispectral raw image data of the reservoir area, remove systematic errors from the multispectral raw image data, and obtain a sharpened image; The edge intensity and texture gradient of the multispectral band are extracted from the sharpened image. If the edge intensity and texture gradient both exceed the preset intensity threshold and gradient threshold, the corresponding pixel is extracted and connected component morphological closing operation is performed to obtain the boundary of the core target region. The extended core region is obtained by statistically analyzing the multispectral pixel distribution histogram within the boundary of the core target region and extending it outward to adjacent pixels. A multispectral saliency map is generated based on the extended core region. The pixel values ​​in the multispectral saliency map are traversed to determine the relevant areas for reservoir dam safety monitoring, and the core protection zone mask is obtained. The extended core region is decomposed into a block-based three-level discrete wavelet transform to generate a sub-band coefficient matrix. The sub-band coefficient matrix is ​​then encoded with a low quantization step size according to the core protection zone mask to obtain core compressed data. A fourth-level discrete wavelet transform is performed on the background regions in the sharpened image that do not belong to the extended core region, and high-quantization step-size encoding is performed to generate background compressed data. The core compressed data and the background compressed data are then fused to obtain a hybrid compressed image. Embedded code block data is extracted from the hybrid compressed image, and a distortion metric is calculated based on the embedded code block data. If the distortion metric is lower than a preset distortion threshold, the hybrid compressed image is confirmed as the final transmission image. At the receiving end, multi-layer resolution decoding is performed on the final transmitted image, and the hybrid compressed image is restored and reconstructed according to the core protection zone mask to obtain a complete image of the reservoir dam and water level safety monitoring.

[0008] Secondly, the present invention provides an image compression and transmission system for a satellite image transmission monitoring station of a reservoir, comprising: The data acquisition module is used to acquire multispectral raw image data of the reservoir area, remove systematic errors from the multispectral raw image data, and obtain a cleared image. The data extraction module is used to extract the edge intensity and texture gradient of the multispectral band from the sharpened image. If the edge intensity and the texture gradient both exceed the preset intensity threshold and gradient threshold, the corresponding pixel points are extracted and connected component morphological closing operation is performed to obtain the boundary of the core target region. The pixel expansion module is used to statistically analyze the multispectral pixel distribution histogram within the boundary of the core target region and expand it outward to adjacent pixels to obtain the expanded core region. The mask generation module is used to generate a multispectral saliency map based on the extended core region, traverse the pixel values ​​in the multispectral saliency map to determine the relevant areas for reservoir dam safety monitoring, and obtain the core protection zone mask. The data decomposition module is used to perform block-based three-level discrete wavelet transform decomposition on the extended core region to generate a sub-band coefficient matrix, and to perform low-quantization step-size encoding on the sub-band coefficient matrix according to the core protection zone mask to obtain core compressed data. The image compression module is used to perform a fourth-level discrete wavelet transform on the background region in the sharpened image that does not belong to the extended core region, and perform high-quantization step-size encoding to generate background compressed data. The core compressed data and the background compressed data are then fused to obtain a hybrid compressed image. The data calculation module is used to extract embedded code block data from the hybrid compressed image, calculate a distortion metric value based on the embedded code block data, and if the distortion metric value is lower than a preset distortion threshold, then the hybrid compressed image is confirmed as the final transmission image. The image decoding module is used to perform multi-layer resolution decoding on the final transmitted image at the receiving end, and to restore and reconstruct the hybrid compressed image according to the core protection zone mask to obtain a complete image of the reservoir dam and water level safety monitoring.

[0009] Compared with the prior art, the present invention has the following beneficial effects: (1) This invention uses a progressive core area locking method to accurately delineate the core area for safety monitoring, such as the dam-water level interface and cracks. Then, by using low-quantization step-size encoding for the core area, the peak signal-to-noise ratio of the core area is improved, the detail distortion is reduced, and the crack width recognition accuracy is improved, thus solving the drawbacks of traditional schemes that focus on compression and neglect details.

[0010] (2) This invention eliminates sensor system errors and terrain projection distortion by radiometric calibration and geometric orthorectification, thereby improving the image signal-to-noise ratio. The core area identification process combines multispectral band characteristics and uses algorithms such as Sobel operator, gray-level co-occurrence matrix, and connected component labeling to accurately locate the core area even in complex riverbanks and vegetation cover, reducing false detection rate and false negative rate. It has better regional identification capabilities than traditional schemes and ensures monitoring stability under different meteorological and terrain conditions.

[0011] (3) This invention uses distortion verification to trigger recompression of images that do not meet the threshold, generating a final transmission image with a verification identifier. The receiving end can quickly verify the integrity of the image. During the decoding stage, the core data is restored without loss based on the core protection zone mask, and the background area is reconstructed using context adaptive interpolation. This ensures that key details are not distorted and eliminates the block effect of background compression, making the complete image after splicing visually coherent and the geographical coordinates accurate. This solves the problem of transmission without verification and decoding distortion in traditional solutions. Attached Figure Description

[0012] Figure 1 This is a schematic diagram of the image compression and transmission method for a satellite image transmission monitoring station for a reservoir provided in the first embodiment of the present invention; Figure 2 This is a schematic diagram of the image compression and transmission system for a satellite image transmission monitoring station used in a reservoir, provided in the second embodiment of the present invention. Detailed Implementation

[0013] The technical solutions of 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 within the scope of protection of the present invention.

[0014] Reference Figure 1 The first embodiment of the present invention provides an image compression and transmission method for a satellite image transmission monitoring station for a reservoir, comprising the following steps: S11, acquire multispectral raw image data of the reservoir area, remove systematic errors from the multispectral raw image data, and obtain a sharpened image; S12, extract the edge intensity and texture gradient of the multispectral band from the sharpened image. If the edge intensity and the texture gradient both exceed the preset intensity threshold and gradient threshold, extract the corresponding pixel points and perform connected component morphological closing operation to obtain the boundary of the core target region. S13, Calculate the multispectral pixel distribution histogram within the boundary of the core target region, and extend it outward to adjacent pixels to obtain the extended core region; S14, Generate a multispectral saliency map based on the extended core region, traverse the pixel values ​​in the multispectral saliency map to determine the relevant areas for reservoir dam safety monitoring, and obtain the core protection zone mask; S15, perform block-based three-level discrete wavelet transform decomposition on the extended core region to generate a sub-band coefficient matrix, and perform low-quantization step-size encoding on the sub-band coefficient matrix according to the core protection zone mask to obtain core compressed data; S16, Perform a fourth-level discrete wavelet transform on the background region in the sharpened image that does not belong to the extended core region, and perform high-quantization step-size encoding to generate background compressed data. Then, fuse the core compressed data with the background compressed data to obtain a hybrid compressed image. S17, extract embedded code block data from the hybrid compressed image, calculate a distortion metric value based on the embedded code block data, and if the distortion metric value is lower than a preset distortion threshold, then confirm that the hybrid compressed image is the final transmission image; S18, at the receiving end, multi-layer resolution decoding is performed on the final transmitted image, and the hybrid compressed image is restored and reconstructed according to the core protection zone mask to obtain a complete image of the reservoir dam body and water level safety monitoring.

[0015] In step S11, acquiring multispectral raw image data of the reservoir area and removing systematic errors from the multispectral raw image data to obtain a sharpened image includes: Extract the sensor response grayscale values ​​from the original multispectral image data, and perform a linear transformation based on the sensor response grayscale values ​​to obtain an absolute radiance image; Calculate the pixel offset of the absolute radiance image. If the pixel offset is greater than a preset offset threshold, resample the absolute radiance image to obtain a resampled pixel array. Orthorectification is performed based on the resampled pixel array to obtain a sharpened image.

[0016] It should be noted that the original multispectral image data comes from satellite remote sensing sensors and includes four core bands: blue, green, red, and near-infrared. Each band is stored in 16-bit unsigned integer format with a grayscale value range of 0 to 65535. Image pixel data is read one by one by band to generate a grayscale value matrix for each band. The band index is as follows: 1 for blue band, 2 for green band, 3 for red band, and 4 for near-infrared band. x and y are pixel coordinates, and the corresponding position is the sensor response grayscale value of the pixel. The sensor response grayscale value and absolute radiance satisfy a linear relationship. In the transformation formula, the sensor response grayscale value of a single band is used as the basis for calculation. First, the grayscale value is multiplied by the radiative response coefficient of the corresponding band, and then the corresponding band's radiative response coefficient is added. The radiation offset, which includes dark current correction values, is the final calculated absolute radiance value of the pixel in the corresponding band, reflecting the true radiation intensity of the ground object. The dark current gray value is the dark current gray value of the band, which was calibrated to 50±5 in the laboratory before satellite launch. The radiation response coefficient is determined based on the satellite's on-orbit calibration data. The radiation offset is obtained by fitting the calibration data and is used to offset the influence of sensor circuit offset to ensure the accuracy of radiance calculation. The gray value matrix of each band is substituted pixel by pixel into the transformation formula to obtain the absolute radiance matrix of each band. The matrix is ​​integrated in the band order of blue, green, red, and near-infrared to generate a three-dimensional data volume of "band-pixel-radiance", that is, the absolute radiance image.

[0017] Clear and stable geographic landmarks were selected from the images, such as the corners of reservoir dams, inflection points of reservoir boundaries, vertices of surrounding fixed buildings, and bridge endpoints. A total of 15-20 evenly distributed landmarks were chosen. The theoretical pixel coordinates of each landmark were obtained using satellite ephemeris data and a Geographic Information System (GIS). The actual pixel coordinates of the landmarks in the absolute radiance image were extracted using the Harris corner detection algorithm. The corner detection threshold was set to 0.04, and the pixel offset of a single landmark was the Euclidean distance between the theoretical and actual coordinates. The average offset of all landmarks was calculated as the global pixel offset. A preliminary threshold range was set by collecting 500 sets of multispectral sharpened images of different reservoirs, under different meteorological conditions, and at different shooting times, referencing the conventional threshold range for Harris corner detection in remote sensing image processing. With a threshold value ranging from 0.03 to 0.05, corner detection was performed on 500 sample images using three candidate thresholds: 0.03, 0.04, and 0.05. When the threshold was 0.03, an average of 320 corners were detected per image, but the false detection rate reached 18%. When the threshold was 0.04, an average of 210 corners were detected per image, the false detection rate dropped to 5%, and the effective corner detection rate reached 96%, completely covering the entire image area. When the threshold was 0.05, an average of 120 corners were detected per image, although the false detection rate was only 3%, the effective corner detection rate dropped to 82%, and there was a problem of missing edge markers. Considering the false detection rate, effective detection rate, and subsequent offset calculation accuracy, the corner detection threshold was determined to be 0.04. Corner points extracted under this threshold can achieve a pixel offset calculation error of ≤0.1 pixels, meeting the accuracy requirements of geometric correction.

[0018] Based on the geometric accuracy requirements of reservoir monitoring, for example, a pixel-level positioning error ≤ 1 pixel, corresponding to an actual error ≤ 10m at a ground resolution of 10m, the preset offset threshold is 1.0 pixel. The offset threshold setting is strongly correlated with the satellite ground resolution. For mainstream water conservancy monitoring satellites with a ground resolution of 10m, the preset offset threshold is 1.0 pixel. For satellites with a resolution of 5m and higher, the threshold is adjusted to 0.8 pixels. If the global pixel offset is ≤ 1.0 pixel, it indicates that the geometric offset is within the allowable range, and no resampling is required. The pixel array of the absolute radiance image is directly used. If the global pixel offset is > 1.0 pixel, resampling correction is performed. For example, the processing logic when the pixel offset is 18 pixels > 1 pixel threshold. The threshold can be fine-tuned according to the ground resolution of the satellite sensor. The offset threshold is set by collecting 100 sets of original satellite images of different reservoirs and calculating their global pixel offset. Statistical results show that 90% of normal images have an offset ≤ 1.0 pixel, and only 10% of images have an offset > 1.0 pixel due to terrain undulations and slight orbital offsets.

[0019] A bilinear interpolation algorithm is employed, which balances correction accuracy and computational efficiency, conforms to the conventional choice for geometric correction of remote sensing images, and avoids the jagged distortion caused by nearest-neighbor interpolation. Based on theoretical pixel coordinates, a regular sampling grid with the same resolution as the original image is constructed, with the grid spacing equal to the ground resolution of the original pixels, such as 10 meters / pixel, adapting to the conventional configuration of satellite sensors. For each node in the sampling grid, the corresponding radiance value is calculated using the bilinear interpolation formula. Taking the node to be interpolated in the sampling grid as the target, the four adjacent pixels of the node in the original image are first found, and the horizontal and vertical distances between the four pixels and the node to be interpolated are calculated respectively. The interpolation weight of each pixel is calculated based on the inverse ratio of distance, ensuring that the sum of the four weights is 1. Then, the radiance values ​​of the four pixels are multiplied by their corresponding weights and summed. The final result is the resampled radiance value of the node to be interpolated. The resampled pixel radiance values ​​are integrated with the interpolation results of all nodes to generate a resampled pixel array of the same size as the original image without geometric offset.

[0020] A 1:10,000 digital elevation model (DEM) of the reservoir area was used, covering the entire image area and a 5-kilometer radius around it. This was used to calculate the projection error caused by topographic relief. Orbital parameters from satellite ephemeris data were obtained, such as orbital altitude of 500-700 km, inclination angle of 98°-99°, and azimuth angle dynamically changing over time. In-sensor azimuth elements, such as principal point coordinates x0 and y0 of 512 pixels and focal length f of 150 mm, were also obtained. The parameter accuracy met the standards for publicly available satellite remote sensing data. Six to eight evenly distributed ground control points were selected in the image, such as the midpoint of the dam crest, the coordinates of reservoir water level stations, and bridge endpoints. The geodetic coordinates of the control points were obtained through GPS measurements, with latitude and longitude accurate to 0.0001° and elevation accurate to 0.1 meters, as correction benchmarks.

[0021] Based on DEM data and sensor parameters, the terrain projection error for each pixel is calculated. Using a single pixel as the calculation unit, the difference between the terrain elevation corresponding to that pixel and the normal reservoir water level elevation is first calculated. Then, the distance from that pixel to the image's principal point is calculated. The elevation difference is multiplied by the distance from the pixel to the principal point, and then divided by the difference between the satellite orbit altitude and the normal reservoir water level elevation. The final result is the terrain projection error value for that pixel. The terrain elevation corresponding to the pixel is taken from DEM data, the normal reservoir water level elevation, the satellite orbit altitude, and the distance from the pixel to the image's principal point. Based on the projection error and ground control points, RPC (with...) is used... The rational polynomial coefficient (RPC) model is used to correct the coordinates of each pixel in the resampled pixel array to obtain the target coordinates under orthorectified projection. The corrected target coordinates are then interpolated again using a bilinear interpolation algorithm to obtain the corresponding radiance value from the DEM data and the resampled pixel array. The RPC model is a general model for orthorectification of satellite remote sensing images. The coefficients of the RPC rational polynomial coefficient model are directly adopted from the RPC coefficients provided in the satellite data product. If there are no built-in coefficients, the 0th to 3rd order RPC coefficients are obtained by fitting the geodetic coordinates and pixel coordinates of 6 to 8 ground control points using the least squares method.

[0022] The orthorectified radiance matrices of each band are integrated in the order of blue, green, red, and near-infrared to generate an orthorectified multispectral image. The ground control point error of the corrected image is calculated, i.e., the root mean square error RMSE ≤ 0.5 pixels, corresponding to a ground error ≤ 5 meters. The image is stored in the standard GeoTIFF format, containing metadata such as geographic coordinate information, radiance attributes, and sensor parameters. The output is a cleared image that has been freed from systematic errors and geometric distortions.

[0023] In step S12, the edge intensity and texture gradient of the multispectral bands are extracted from the sharpened image. If the edge intensity and the texture gradient both exceed a preset intensity threshold and a preset gradient threshold, the corresponding pixels are extracted and connected component morphological closing operations are performed to obtain the boundary of the core target region, including: Multispectral bands are extracted from the sharpened image, a Sobel template is applied to the multispectral bands, the gradient magnitudes in the horizontal and vertical directions are calculated, and the edge intensity of the multispectral bands is synthesized. Local windows in the sharpened image are located based on the edge intensity, and the local windows are processed by a preset gray-level co-occurrence matrix to obtain the texture gradient; If the edge intensity and the texture gradient both exceed the preset intensity threshold and gradient threshold, then the corresponding pixel points are extracted to obtain a set of high-frequency abrupt pixel points; Perform connected component morphological closing operations on the set of high-frequency abruptly changing pixels to obtain the boundary of the core target region.

[0024] It should be noted that the core monitoring bands extracted from the sharpened image are the visible red band and the near-infrared band. After calibration for reservoir monitoring scenarios, these two bands have the highest distinguishability between water bodies, dam bodies, and bank slopes, and the most significant edge features. Each band is a single-channel grayscale image with a pixel value range of 0 to 255. The pixel value range is obtained by normalizing the absolute radiance image, and the image resolution is consistent with the sharpened image. The Sobel template uses a 3×3 standard template. The horizontal template is used to detect horizontal edges, and the vertical template is used to detect vertical edges. For each pixel of each band, convolution operations are performed with the two templates to obtain the horizontal gradient and the vertical gradient. The gradient magnitude of a single pixel is the square root of the sum of the squares of the horizontal gradient and the squares of the vertical gradient.

[0025] When synthesizing edge intensity, the red band has a weight of 0.6 and the near-infrared band has a weight of 0.4. This weight is based on the measured statistics of 500 sets of clear satellite images of the reservoir. The red band contributes 62% to the identification of core edges such as dam cracks and dam surface erosion; the near-infrared band contributes only 35% to the identification of internal cracks in the dam. The red band is less affected by vegetation interference and can still stably identify the edges of the dam concrete. The near-infrared band can penetrate thin fog and supplement the edge signal of the water-bank boundary. The weight ratio of 0.6:0.4 can balance the identification needs of the two types of scenes. The synthesis formula is the red band gradient amplitude multiplied by 0.6 plus the near-infrared band gradient amplitude multiplied by 0.4. The final multispectral band edge intensity value ranges from 0 to 255. The larger the value, the more significant the edge. Set the edge strength threshold to 120. Center the pixels with an edge strength of not less than 120 and expand outwards to form a 64×64 pixel square local window. If the window overlap rate of adjacent trigger pixels exceeds 50%, they are merged into one window. The final number of generated local windows is controlled between 5 and 20. The edge strength threshold was statistically set based on the edge strength of 500 sets of cleared reservoir images. The results showed that the edge strength values ​​at the junction of the dam body and water level, and at the bank slope edges, were concentrated between 120 and 210; the edge strength values ​​in vegetated areas and uniform water areas were concentrated between 0 and 60; and the edge strength values ​​at the junction of vegetation and bank slopes were concentrated between 60 and 120. The filtering effects of thresholds 100, 120, and 140 were tested. At a threshold of 100, the average number of local windows was 35, including a large number of transition areas, resulting in high computational redundancy. At a threshold of 120, the average number of local windows was 12, covering only effective edge areas, improving computational efficiency by 65%, and eliminating any missed effective edges. At a threshold of 140, the average number of local windows was 4, but there were some missed edges at narrow dam bodies. Considering both filtering accuracy and computational efficiency, an edge strength threshold of 120 was determined.

[0026] Gray-level compression is performed on each local window, compressing pixel values ​​from 0-255 to 16 gray levels (0-15). Within each window, a 16×16 dimensional gray-level co-occurrence matrix is ​​calculated with a pixel spacing of 1 and orientations of 0°, 45°, 90°, and 135°. This matrix represents the probability of gray level i and gray level j occurring simultaneously at corresponding orientations and spacings. Three core features—contrast, correlation, and entropy—are extracted from the co-occurrence matrix. Contrast reflects the clarity of the texture. By traversing all gray value combinations in the gray-level co-occurrence matrix, the squared difference between two gray values ​​is multiplied by the value of that gray value combination in the matrix. Contrast is the probability of occurrence of each gray value. The sum of all the results is the contrast. Correlation reflects the consistency of the texture. It is calculated by iterating through all gray value combinations in the gray-level co-occurrence matrix, subtracting the mean gray value from each of the two gray values, multiplying the result by the probability of occurrence of the gray value combination, summing all the results, and then dividing by the product of the standard deviations of the two gray values. The final result is the correlation. Entropy reflects the complexity of the texture. It is calculated by iterating through all non-zero probability gray value combinations in the gray-level co-occurrence matrix, multiplying the probability of occurrence of the gray value combination by the base-2 logarithm of that probability, summing all the results, and taking the negative number. The final result is the entropy. The texture gradient is calculated by multiplying contrast by 0.5, subtracting the absolute value of correlation by 0.3, and then adding entropy by 0.2. Based on 500 sets of experimental data, this weighting shows that contrast contributes 55% to the recognition of core textures, far exceeding the other two features. Therefore, it is assigned the highest weight of 0.5. The contribution of correlation is only 25%, and the recognition contribution of entropy is only 20%. When the weights are 0.5, 0.3, and 0.2, the discrimination accuracy is improved to 94%. The gradient values ​​of core textures such as dam cracks and water level lines are more than 3 times higher than those of background textures. The texture gradient of each local window is assigned to all pixels within the window, and the texture gradient of non-window areas is set to 0, generating a texture gradient map of the same size as the edge intensity map.

[0027] It should be noted that the parameters of the gray-level co-occurrence matrix mentioned above were calculated by selecting manually labeled core and non-core regions from 500 sets of reservoir remote sensing images, and calculating texture features at intervals of 1, 2, and 3, and directions of 0°, 45°, 90°, and 135°, respectively. The texture gradient has the highest distinguishability of the core region when the interval is 1, and the distinguishability decreases significantly as the interval increases. The four directions cover the horizontal, vertical, and diagonal directions, which can completely describe the anisotropy of the texture. The gray level is compressed to 16 levels, which retains texture information while keeping the computational complexity within an acceptable range. The classification accuracy has been verified to be 12% higher than that of 8 levels, and only 2% lower than that of 32 levels, while the computational load is reduced by half.

[0028] The preset intensity threshold is 150, and the gradient threshold is 1.9. For each pixel in the image, if both the edge intensity and texture gradient are not lower than the intensity threshold, it is identified as a high-frequency mutation pixel and included in the high-frequency mutation pixel set. This set uses pixel coordinates as elements. The intensity threshold is determined by statistically analyzing the edge intensity distribution of effective core edges in 500 samples. 92% of the effective edge intensity values ​​are ≥150, and 8% of the noise edge intensity values ​​are <150. Setting 150 as the intensity threshold ensures that the recognition rate of effective edges is ≥92%, and the false detection rate of noise edges is ≤8%, thus ensuring high-frequency mutation pixels... The purity of the element set; the gradient threshold was determined by statistically analyzing the gradient distribution of different textures in the sample. The gradient values ​​of the dam and water body boundary textures were concentrated between 1.9 and 2.8, while the gradient values ​​of vegetation textures and water ripples were concentrated between 0.5 and 1.8. Three candidate thresholds of 1.7, 1.9, and 2.1 were tested. When the threshold was 1.7, the texture false detection rate reached 12%; when the threshold was 1.9, the texture false detection rate dropped to 4%, and the effective boundary texture detection rate reached 95%; when the threshold was 2.1, the effective texture detection rate dropped to 88%. The gradient threshold of 1.9 was determined to achieve accurate differentiation between effective boundary textures and noise textures.

[0029] Morphological closing operations employ a 3×3 square structuring element. The closing operation logic is dilation followed by erosion, i.e., dilation is performed on the set of high-frequency abrupt pixel changes to fill the pores, and then erosion is performed to restore the true size of the boundary. The pixel set after the closing operation is labeled with connected components using the 8-connectivity criterion, resulting in multiple connected components. The minimum connected component area threshold is set to a two-level adaptation system. The primary threshold of 50 pixels corresponds to the identification of the main boundary area between the dam body and the water level line, ensuring the locking of the core monitoring zone. For minor abnormal areas such as dam body cracks and local seepage, a secondary minimum connected component threshold of 5 pixels is set simultaneously. For the retained maximum connected component, the Canny edge detection algorithm is used to extract the outer boundary, resulting in the boundary of the continuously closed core target area. The 8-connectivity criterion means that pixels in eight adjacent directions are considered connected. The maximum connected component corresponds to the main boundary area between the dam body and the water level line.

[0030] In step S13, the step of statistically analyzing the multispectral pixel distribution histogram within the boundary of the core target region and expanding it outward to adjacent pixels to obtain the extended core region includes: Extract the multispectral pixel values ​​within the boundary of the core target region, and construct a multispectral pixel distribution histogram based on the multispectral pixel values; Statistical feature parameters are calculated based on the pixel distribution histogram of the multispectral array, and the similarity between the statistical feature parameters and adjacent pixels is quantified to obtain a similarity metric. The extended range is obtained by performing region growing based on the similarity metric. The extended area is then filled with holes to obtain the extended core region.

[0031] It should be noted that, using the boundary of the closed core target area output by S12 as the spatial range constraint, the multispectral values ​​of all pixels within the boundary are extracted band by band from the sharpened image output by S11, including four core bands: red, green, blue, and near-infrared, which are completely consistent with the previous steps. The four-dimensional band value vector of each pixel is used as an independent sampling point, and the distribution frequency of the brightness value of each band in the range of 0~255 is statistically analyzed to construct a multispectral joint pixel value distribution histogram with a dimension of 4×256. This histogram takes into account both the independent distribution characteristics of each band and the joint correlation characteristics between multiple bands, and can accurately characterize the spectral patterns of pixels within the boundary. Among them, pixels near the water level line will show a bimodal distribution characteristic with lower near-infrared values ​​and higher red band values, while the crack area of ​​the dam body will show a local trough distribution characteristic of blue and green bands.

[0032] Based on the constructed multispectral pixel distribution histogram, 22 statistical feature parameters that comprehensively characterize the spectral distribution of pixels within the boundary are calculated. These include the mean, variance, skewness, and kurtosis of each of the four bands (red, green, blue, and near-infrared), as well as six covariance matrix parameters formed by pairwise combinations of the four bands. Using the 22 statistical feature parameters of each pixel within the boundary as a reference vector, the Euclidean distance between this pixel and the statistical feature parameter vectors of each of its eight neighboring pixels is calculated. Then, by applying a preset maximum distance threshold of 32, the similarity metric of adjacent pixels is quantized to obtain the similarity measure value. The baseline within the boundary is then calculated. The similarity measure between two pixels is calculated by subtracting the ratio of this Euclidean distance to a preset maximum distance threshold from the 22-dimensional statistical feature vectors of the quasi-pixel and its neighboring pixels. Specifically, the 22-dimensional statistical feature parameters of pixels within the boundary of the core target region are extracted from 500 samples, and the Euclidean distance between pixels within the boundary and their neighboring pixels is calculated. Statistical results show that the Euclidean distance between similar pixels is ≤18, and the Euclidean distance between dissimilar pixels is ≥32. Setting the maximum distance threshold to 32 ensures that the similarity measure of similar pixels is ≥0.4375, and the similarity measure of dissimilar pixels is ≤0.

[0033] An initial seed point is set for the region growth. From all pixels on the boundary of the core target region, the pixel with the highest similarity metric value and a similarity metric value of not less than 0.92 is selected as the initial growth seed point to ensure that the spectral characteristics of the growth starting point are highly matched with the core boundary region. Centered on the initial growth seed point, the region expands outward using an eight-neighbor iterative growth rule. In each iteration, only pixels with a similarity metric value of not less than 0.75 in the eight-neighbor region are included in the growth range. At the same time, the newly included pixels are used as the growth seed point for the next iteration. The iteration terminates when the number of newly added pixels in two consecutive iterations does not exceed 5, avoiding meaningless over-expansion that leads to redundancy in the core region, and finally forming a continuous initial expansion range.

[0034] The similarity measurement threshold is determined by statistically analyzing the distribution of similarity measurement values ​​between pixels within the core region boundary and adjacent pixels in 500 sample groups. The similarity measurement value between pixels within the boundary and adjacent pixels belonging to the core region is not lower than 0.92, while the similarity measurement value between pixels within the boundary and adjacent pixels in non-core regions is lower than 0.75. Therefore, setting the seed point screening threshold to 0.92 and the region growth threshold to 0.75 can ensure that the growth starting point is accurate and that the expansion process does not mistakenly enter non-core regions.

[0035] A morphological filling algorithm based on the 8-connectivity criterion is used to fill holes in the initial expanded region obtained from region growing. A hole diameter threshold of 15 pixels is initially set, and only isolated holes with a diameter not exceeding 15 pixels within the expanded region are filled. Holes exceeding this threshold are identified as gaps in real features such as dam drainage holes or reservoir bank rock fissures and are not filled. The filling process is performed pixel-by-pixel from the hole boundary inwards, filling isolated holes within the expanded region caused by local noise and low-similarity pixels, ensuring that the expanded region forms a continuous, solid, and complete area. The final result is an extended core area that fully covers the traces of water level fluctuations and the characteristics of dam cracks. Among the 500 samples, the diameter of the cavities in the extended area was statistically analyzed. Noise cavities (caused by local noise and low similarity pixels) had a diameter ≤15 pixels, accounting for 95%; the diameter of real feature gaps (dam drainage holes, reservoir bank rock gaps) had a diameter ≥20 pixels, accounting for 5%. The cavity diameter threshold was set to 15 pixels, and only cavities ≤15 pixels were filled to avoid filling real feature gaps and to fill noise cavities, thus ensuring the continuity of the extended core area.

[0036] In step S14, the process of generating a multispectral saliency map based on the extended core region, traversing the pixel values ​​in the multispectral saliency map to determine the relevant areas for reservoir dam safety monitoring, and obtaining the core protection zone mask includes: Multidimensional local contrast features are calculated based on the multispectral reflectance data within the extended core region, and a multispectral saliency map is generated based on the multidimensional local contrast features. The pixel values ​​of the multispectral saliency map are traversed. If a pixel value is greater than a preset adaptive threshold, it is marked as a high saliency score pixel. Connectivity component labeling is performed on the highly significant fractional pixels to obtain the relevant area for reservoir dam safety monitoring; A binary image matrix is ​​generated based on the relevant area of ​​the reservoir dam safety monitoring. The elements in the binary image matrix are assigned mask identifiers to obtain the core protection area mask.

[0037] It should be noted that a 3×3 local analysis window is constructed centered on each pixel within the extended core region. The local contrast of each of the four bands is calculated. First, the absolute difference between the reflectance of the target pixel in the corresponding band and the average reflectance of that band within the window is calculated. Then, this difference is divided by the sum of the average reflectance within the window and the minimum value of 0.001, thus avoiding calculation errors where the denominator is 0. Considering the band sensitivity of the reservoir monitoring scenario, based on 500 sets of multispectral sharpened images of different reservoirs and scenarios, the contribution of each band to the identification of the core region is statistically analyzed, and this weight is set accordingly. Statistical data shows that the red band and near-infrared band contribute 31% and 30% respectively to the identification of the core region, which are key factors for core region identification. The green and blue bands, at 20% and 19% respectively, primarily serve to assist in eliminating interference. Weights are set at 0.3 for the red band, 0.2 for the green band, 0.2 for the blue band, and 0.3 for the near-infrared band. The local contrast of each of the four bands is weighted and fused to obtain the multidimensional local contrast features of each pixel. A higher value indicates a more significant difference in reflectivity between the pixel and its surrounding area, suggesting a potentially critical area related to dam safety. Linear normalization is applied to the multidimensional local contrast features of all pixels within the extended core area, mapping them to the 0-1 interval to obtain a saliency score for each pixel. This generates a single-channel grayscale image of the same size and coordinate system as the sharpened image, i.e., a multispectral saliency map.

[0038] The Otsu method, a common approach in image processing, is used to calculate an adaptive threshold. By traversing all possible thresholds within the 0-1 range, pixels with saliency scores are divided into foreground regions (above the threshold) and background regions (below the threshold). The pixel proportions of the foreground and background regions, as well as the mean saliency scores of the two regions, are calculated separately. The inter-class variance corresponding to the threshold is obtained by multiplying the pixel proportions of the foreground and background regions by the square of the difference between the two region means. Finally, the threshold corresponding to the maximum inter-class variance is determined as the Otsu optimal adaptive threshold. A threshold offset of 0.1 is added simultaneously to improve the screening strictness, retaining only the highly saliency regions with the strongest correlation to dam safety. All pixels in the extended core region within the multispectral saliency map are traversed. For each pixel, a judgment is performed. If the saliency score of the pixel is greater than the calculated adaptive threshold, it is marked as a highly saliency pixel; otherwise, it is marked as a non-saliency pixel. Finally, a binary label matrix of the same size as the saliency map is generated, where highly saliency pixels are assigned a value of 1, and non-saliency pixels are assigned a value of 0.

[0039] It is worth noting that the threshold offset of 0.1 was determined by selecting 500 sets of satellite remote sensing images of the reservoir, calculating the optimal Otsu threshold for each set of images, and using manually marked core protected areas as a benchmark. The impact of the threshold offset within the range of 0.05 to 0.15 on the accuracy and false detection rate of core area identification was tested. Statistical results show that when the offset is 0.10, the average accuracy of core area identification reaches 96.5%, and the false detection rate is 3.2%, exhibiting the best overall performance. When the offset is below 0.08, the false detection rate rises to over 6%. When the offset is above 0.12, the accuracy drops to below 91%. Therefore, the offset is set to 0.10 to improve the screening rigor while ensuring the integrity of the core area.

[0040] Generate a binarized image matrix with the same size, coordinate system, and projection information as the sharpened image output in step S11. The matrix dimensions are completely consistent with the sharpened image, and all initial values ​​of the matrix are set to 0. Then, traverse all pixel coordinates of the relevant area of ​​the reservoir dam safety monitoring, and set the element at the corresponding coordinate position in the matrix to 1. Finally, a standard binarized image matrix is ​​formed where the relevant area of ​​safety monitoring is 1 and the rest of the area is 0. Define the value 1 in the binarized image matrix as a mask identifier, representing the core protection area that needs to be encoded with low quantization step size and high fidelity, and define the value 0 as a non-core area identifier, representing the background area that can be encoded with high quantization step size and high compression ratio. This binarized matrix is ​​the core protection area mask.

[0041] In step S15, the extended core region is subjected to block-based three-level discrete wavelet transform decomposition to generate a sub-band coefficient matrix. The sub-band coefficient matrix is ​​then encoded with a low-quantization step size based on the core protection zone mask to obtain core compressed data, including: Based on the core protection zone mask, a block-based three-level discrete wavelet transform decomposition is performed on the extended core region to generate a sub-band coefficient matrix; The core protection zone mask is mapped to a pre-constructed transform domain coordinate system to obtain the core coefficient index set corresponding to the sub-band coefficient matrix; If the coordinates of the sub-band coefficient matrix belong to the core coefficient index set, then the coefficients corresponding to the coordinates of the sub-band coefficient matrix are quantized to obtain high-precision quantized coefficients. The high-precision quantization coefficients are encoded to generate a compressed bitstream, and the compressed bitstream is encapsulated to obtain core compressed data.

[0042] It should be noted that the extended core region is the multispectral image output by S13, and the core protection area mask is the binarized matrix output by S14. Block processing is performed, using a fixed size of 64×64 pixels adapted to the JPEG2000 encoding standard. The extended core region is divided into non-overlapping blocks, and the block boundaries are completely aligned with the pixel grid of the original image. If there are blocks with less than 64×64 pixels at the edge of the extended core region, they are filled to the standard block size using a mirror extension method. The mirror extension uses bidirectional symmetrical extension in both horizontal and vertical directions. If the width of the block at the edge of the extended core region is less than 64 pixels, it is mirrored to the left with the right boundary of the block as the axis of symmetry and extended to 64 pixels; if the height is less than 64 pixels, it is mirrored upward with the lower boundary of the block as the axis of symmetry and extended to 64 pixels. The pixel values ​​after extension are completely consistent with the original pixel values ​​at the symmetrical positions. Discrete wavelet transform decomposition is performed, and a three-level decomposition is carried out on each image block using the Daubechies9 / 7 bioorthogonal wavelet basis. This wavelet basis is the default lossy compression wavelet basis of the JPEG2000 standard, which achieves the optimal balance between edge preservation and energy compression performance in reservoir remote sensing images.

[0043] The specific process of the three-level decomposition is as follows: The first-level decomposition decomposes each block image into one low-frequency approximation sub-band LL1 and three high-frequency detail sub-bands LH1, HL1, and HH1 (horizontal, vertical, and diagonal). The second-level decomposition further decomposes the first-level low-frequency sub-band LL1 to generate the second-level low-frequency sub-band LL2 and three corresponding high-frequency sub-bands. The third-level decomposition performs the final decomposition on the second-level low-frequency sub-band LL2 to obtain the third-level low-frequency approximation sub-band LL3 and three corresponding high-frequency sub-bands. Among them, the low-frequency sub-band LL3 retains the overall structural information of the dam body and water level line, and the high-frequency sub-bands at each level capture detailed edge information such as dam body cracks, water level fluctuation traces, and leakage anomalies at different resolutions, which are completely matched with the core features of reservoir safety monitoring. Finally, all sub-band coefficients of each level of the block are integrated and arranged in order according to band, decomposition level and sub-band direction to generate a sub-band coefficient matrix that corresponds one-to-one with the spatial coordinates of the original extended core region. The matrix contains coefficient data of 4 bands, 1 third-level low-frequency sub-band and 9 high-frequency sub-bands of each level. Each coefficient retains its corresponding spatial domain pixel coordinate mapping relationship.

[0044] A coordinate mapping rule from the spatial domain to the transform domain is established, strictly adhering to the downsampling characteristics of the discrete wavelet transform. One coefficient of the third-level low-frequency sub-band LL3 corresponds to an 8×8 pixel block in the spatial domain, one coefficient of the second-level sub-band corresponds to a 4×4 pixel block in the spatial domain, and one coefficient of the first-level sub-band corresponds to a 2×2 pixel block in the spatial domain, ensuring complete alignment of the physical positions in the spatial domain and the transform domain. Subsequently, mask mapping is performed. First, the core protection zone mask in the spatial domain is fully traversed to identify the coordinate sets of all core protection zone pixels with a mask identifier of 1. Then, according to the resolution mapping rule described above, the core pixel coordinates in the spatial domain are mapped level by level to the transform domain coordinate system of each sub-band.

[0045] The core coefficient determination rule is set as follows: if more than 50% of the pixels in the spatial domain pixel block corresponding to a certain transform domain coefficient belong to the core protection zone, then the coefficient is marked as a core coefficient. This rule can avoid missing effective coefficients at the edge of the core region, while eliminating invalid coefficients that only have a small number of pixels falling into the core region. Finally, all marked core coefficients are integrated to generate a core coefficient index set containing information such as the waveband, sub-band level, sub-band direction, and matrix coordinates of the coefficient. The 50% determination ratio is determined based on the correspondence between wavelet transform coefficients and spatial domain pixel blocks. Through testing at four ratios of 30%, 40%, 50%, and 60%, 50% achieves the optimal balance between reconstruction quality and compression efficiency in the core region, with a peak signal-to-noise ratio loss of less than 0.5dB in the core region and a background bitrate saving of approximately 18%.

[0046] Non-core coefficients were subjected to conventional coarse quantization using the JPEG2000 standard scalar quantization method with dead zones. First, a differentiated quantization step size was set, with a preset baseline quantization step size of 32 for the background region. This was verified through field measurements using 500 sets of satellite remote sensing images of different types of reservoirs under varying meteorological conditions. At a step size of 32, the peak signal-to-noise ratio (PSNR) of the background region was ≥38dB, with no visual distortion, meeting the minimum requirements for reservoir monitoring background interpretation. The overall compression efficiency of a step size of 32 was 30% higher than that of a step size of 24, and the distortion rate was reduced by 60% compared to a step size of 48. Finally, through field calibration, 32 was determined to be the optimal baseline quantization step size for the background region. The quantization step size of the core coefficients is set differently depending on the sub-band type. The low-frequency approximation sub-band LL3 determines the overall structure of the image and requires the highest quantization accuracy. The quantization step size is set to 3, which is about 1 / 10 of the baseline step size. The high-frequency detail sub-bands in the horizontal and vertical directions contain linear key features such as dam cracks and water level lines and require high accuracy. The quantization step size is set to 4, which is 1 / 8 of the baseline step size. The high-frequency sub-bands in the diagonal direction are mostly texture noise, so the quantization step size can be appropriately relaxed and set to 6, which is 1 / 5 of the baseline step size. This differentiated setting takes into account compression efficiency while ensuring detail fidelity.

[0047] Quantization is then performed. For the original wavelet coefficients, their absolute values ​​are first taken, and then the dead zone threshold corresponding to half the quantization step size is subtracted. If the result is less than 0, the quantized coefficient value is directly set to 0. If the result is greater than or equal to 0, the result is divided by the corresponding quantization step size, the integer value is taken, and the sign of the original wavelet coefficient is retained to obtain the quantized coefficient value. For non-core coefficients that do not belong to the core coefficient index set, a standard quantization step size of 32 is uniformly used for regular quantization. Finally, all coefficients in the subband coefficient matrix are traversed, and corresponding coordinates are matched according to the core coefficient index set. Core coefficients that are successfully matched are quantized using the aforementioned low quantization step size, while non-core coefficients are quantized using the standard step size, ultimately generating a high-precision quantized coefficient matrix.

[0048] Embedded encoding is performed using the EBCOT embedded block encoding optimization truncation algorithm of the JPEG2000 standard. The high-precision quantization coefficients are divided into 64×64 blocks for embedded block encoding. The quantization coefficients of each block are bit-plane encoded, and then entropy encoding is performed using adaptive arithmetic encoding to generate independently decodeable embedded compressed blocks.

[0049] Subsequently, a compressed bitstream is generated. All encoded embedded code blocks are sequentially concatenated according to band, sub-band level, and code block position. Simultaneously, complete encoding parameters are written to the bitstream header, including wavelet basis type, decomposition level, quantization step size for each sub-band, and coordinate mapping parameters for the core coefficient index set, generating a continuous compressed bitstream. Finally, data encapsulation is performed, standardizing the compressed bitstream. The encapsulation content includes the main body of the compressed bitstream, spatial coordinate information of the core protection zone mask, code block check codes, and band correspondences. The encapsulation format is fully compatible with the JPEG2000 international standard, resulting in encapsulated core compressed data.

[0050] In step S16, performing a fourth-level discrete wavelet transform on the background regions in the sharpened image that do not belong to the extended core region, and performing high-quantization step-size encoding to generate background compressed data, and fusing the core compressed data with the background compressed data to obtain a hybrid compressed image, includes: A fourth-level discrete wavelet transform is performed on the background regions in the sharpened image that do not belong to the extended core region, and high-quantization step-size encoding is performed to generate background compressed data. Calculate the rate-distortion slope based on the background compressed data, sort the coding blocks corresponding to the rate-distortion slope from high to low to construct a sorting table, and obtain a global sorting table. Traverse the global sorting table to find the optimal cutoff point and obtain the optimal cutoff point position; The core compressed data and the background compressed data are truncated and spliced ​​together according to the optimal truncation point to obtain a hybrid compressed image.

[0051] It should be noted that, based on the cleared image after geometric and radiometric correction output in step S11, the core protection zone mask output in step S14 is used to divide the region. The areas marked as 0 in the mask, excluding the extended core area, are designated as the background area. This area mainly includes non-core content unrelated to the safety monitoring of the reservoir dam, such as uniform water bodies far from the dam, vegetation around the reservoir, distant mountains, and the sky. Using the same block division rules as in step S15, the background area is divided into 64×64 pixel non-overlapping blocks. Blocks with edges less than 64×64 pixels are supplemented to the standard size using a mirror extension method.

[0052] Each background region is divided into blocks and a fourth-level discrete wavelet transform decomposition is performed using the Daubechies9 / 7 biorthogonal wavelet basis. Compared with the third-level decomposition of the core region, this increases the decomposition level by one level, further improving the compression efficiency of the background region. After decomposition, one fourth-level low-frequency approximation sub-band and 12 high-frequency detail sub-bands of three levels (horizontal, vertical, and diagonal) are obtained. The low-frequency sub-band retains only the overall contour information of the background region, while the high-frequency sub-bands correspond to the texture details of the background region.

[0053] The decomposed subband coefficients are subjected to scalar quantization with a high quantization step size. The baseline quantization step size for the background region is 32, and the step size is adjusted according to the subband type. Specifically, the quantization step size for low-frequency approximation subbands is set to 32, the quantization step size for high-frequency subbands in the horizontal and vertical directions is set to 48, and the quantization step size for high-frequency subbands in the diagonal direction is set to 64. By significantly increasing the quantization step size, the amount of invalid data in the background region is compressed. The quantized background region coefficients are processed using the EBCOT embedded block coding optimization truncation algorithm of the JPEG2000 standard. Bit-plane coding is performed in 64×64 blocks, and then entropy coding is completed through adaptive arithmetic coding to generate independent and truncate background compressed data for each background coding block. At the same time, coding parameters such as wavelet decomposition level, quantization step size of each subband, and coding block coordinate mapping relationship are written into the bitstream header.

[0054] Rate-distortion slope is used to characterize the contribution of each unit increase in bit rate of each coding block to the overall image quality improvement. The higher the slope value, the more significant the benefit of the bit rate investment in that coding block to the image quality improvement, and it is the core basis for bit rate allocation. First, the rate-distortion slope is calculated for each core coding block in the core compressed data output in step S15 and each background coding block in the background compressed data. The rate-distortion slope adopts the general calculation formula of the JPEG2000 coding standard. The rate-distortion slope of a single coding block is calculated by the amount of distortion reduction brought by each unit increase in bit rate at the corresponding truncation point. The absolute value of the distortion reduction is divided by the bit rate increase, and the final result is the rate-distortion slope of that coding block. The rate-distortion slopes of all core coding blocks and background coding blocks are normalized. Then, all coding blocks are globally sorted in descending order of rate-distortion slope to construct a global sorting table. The sorting table synchronously records key information such as the region to which each coding block belongs, bit stream address, cumulative bit rate, and corresponding distortion.

[0055] It should be noted that in the above rate-distortion slope calculation, the distortion metric uses the mean square error between the original wavelet coefficients and the quantized wavelet coefficients of the coding block. For each coding block, the cumulative bit rate and cumulative mean square error corresponding to each truncation point are recorded during the encoding process. The absolute value of the decrease in mean square error between adjacent truncation points divided by the increase in bit rate is the rate-distortion slope of the coding block at that truncation point. The rate-distortion slope of each truncation point is obtained by parsing the coding block header information, and the maximum value of the rate-distortion slope in each coding block is taken for global sorting. This distortion metric method is consistent with the rate-distortion optimization truncation algorithm in the JPEG2000 standard.

[0056] The transmission bandwidth constraints of the satellite image transmission monitoring station and the bandwidth constraints of the satellite communication link are obtained. The maximum allowable data volume of a single frame image is used as the core indicator. Based on the actual measurement and calibration of the narrowband satellite communication link commonly used in water conservancy monitoring, the maximum data volume of a single frame image is 500kB, which corresponds to the transmission of a single frame image within 1 second at a bit rate of 500kbps, adapting to the real-time requirements of satellite communication. Then, according to the sorting order of the coding blocks in the global sorting table, starting from the coding block with the highest rate-distortion slope, the bit rate data of each coding block is accumulated sequentially, and the overall image distortion corresponding to the accumulated bit rate is calculated synchronously. During the traversal, a dual constraint condition is set. The first constraint is the core area coding block mandatory inclusion rule, that is, all core area coding blocks must be included in the effective bit stream range. The second constraint is the bandwidth upper limit rule. After all core area coding blocks are included, the bit rate of the background area coding blocks continues to be accumulated until the accumulated bit rate is about to exceed the preset bandwidth constraint upper limit, at which point the traversal stops. This stopping position is the optimal truncation point position.

[0057] Based on the optimal truncation point, the background compressed data is truncated, retaining the background coded block bitstream before the optimal truncation point in the global sorting table and discarding the background coded block bitstream after the truncation point, which has minimal benefit to image quality improvement, thus optimizing the background compressed data. Subsequently, the complete bitstream of the core compressed data and the truncated background compressed bitstream are standardized and spliced ​​according to the bitstream format of the JPEG2000 international standard. During the splicing process, the spatial coordinate order of the coded blocks is strictly followed to ensure that the spatial mapping relationship of the bitstream is completely consistent with the original image. Complete image metadata is written to the header of the spliced ​​bitstream, including image size, band information, wavelet decomposition parameters, quantization step size table, core protection zone mask coordinate information, optimal truncation point parameters, and bandwidth constraint values. Cyclic redundancy check codes are written to the tail of the bitstream for the receiver to verify the integrity of the bitstream. Finally, a hybrid compressed image compatible with the JPEG2000 standard is generated.

[0058] In step S17, the step of extracting embedded code block data from the hybrid compressed image, calculating a distortion metric based on the embedded code block data, and confirming the hybrid compressed image as the final transmission image if the distortion metric is lower than a preset distortion threshold includes: Embedded code block data is extracted from the hybrid compressed image, and the embedded code block data is inversely transformed to generate a locally reconstructed pixel matrix; The difference feature matrix is ​​obtained by performing a difference calculation between the locally reconstructed pixel matrix and the original multispectral image data; The distortion degree is calculated based on the difference feature matrix to obtain the distortion metric value; If the distortion metric is lower than a preset distortion threshold, a verification pass identifier is generated and written into the hybrid compressed image to obtain the final transmitted image.

[0059] It should be noted that, taking the hybrid compressed image output from S16 as input, the encoding parameters in the bitstream header are first parsed, including wavelet decomposition levels, quantization step size table, core protection zone mask coordinate mapping information, and code block segmentation rules. Then, according to the encoding block address index in the bitstream, the embedded code block data corresponding to the core protection zone and background area is extracted block by block. This code block data is an independent decodeable bitstream unit generated by EBCOT embedded block encoding in S15 and S16. The block size is consistent with the encoding stage, which is 64×64 pixels. For each extracted embedded code block data, adaptive arithmetic entropy decoding is first performed to restore the quantized wavelet coefficient matrix. Then, the wavelet coefficient matrix is ​​subjected to the inverse discrete wavelet transform corresponding to the encoding stage. The core protection zone code block is inversely quantized using the low quantization step size of the encoding stage, and the background area code block is inversely quantized using the corresponding high quantization step size. Finally, according to the spatial coordinate order of the original image, the pixel data after the inverse transformation of all code blocks are stitched together to generate a locally reconstructed pixel matrix with the same size, band, and coordinate system as the sharpened image output from S11.

[0060] The sharpened image output by S11, after geometric and radiometric correction, is used as the original reference benchmark for distortion calculation. This image is completely consistent with the source, spatial resolution, band range, and geographic coordinate system of the hybrid compressed image, and has no compression distortion. It is the original true value data for reservoir monitoring. For the local reconstructed pixel matrix and the original sharpened image, pixel-by-pixel difference operation is performed band by band according to the four bands of red, green, blue, and near-infrared. That is, for two pixels with the same spatial coordinates and the same band, the gray value of the reconstructed pixel is subtracted from the gray value of the original reference pixel to obtain the pixel difference at that location. The differences of all bands and all pixels are arranged according to the dimensions of the original image to form a difference feature matrix.

[0061] To meet the operational needs of reservoir dam safety monitoring, a dual-indicator approach is adopted: Peak Signal-to-Noise Ratio (PSNR) weighted by the core region and Structural Similarity (SSIM). PSNR measures pixel-level grayscale deviation, while SSIM measures the preservation of visual features such as image structure and texture. Both indicators are universal standards for evaluating the compression quality of remote sensing images. First, the PSNR is calculated by calculating the mean square error (MSE) of the entire image based on the difference feature matrix. The MSE is the sum of the squares of the differences between all pixels in the difference feature matrix divided by the total number of pixels. Then, the PSNR value of the entire image is calculated using the PSNR formula. This involves calculating the square of the maximum grayscale value of each pixel, dividing it by the MSE, taking the logarithm to base 10, and multiplying the result by 10. The final result is the PSNR, expressed in dB. A higher PSNR value indicates less image distortion and better image quality.

[0062] Structural similarity (SSIM) is calculated by calculating the mean, variance, and covariance of brightness within a local window in both the original and reconstructed images. The product of the two means is multiplied by 2 and a minimum constant is added to form the first term of the numerator. Similarly, the product of the two variances is multiplied by 2 and a minimum constant is added to form the second term of the numerator. The sum of the squares of the two means, plus the minimum constant, forms the first term of the denominator. The sum of the two variances, plus the minimum constant, forms the second term of the denominator. Finally, the product of the two numerators and the result of the product of the two denominators is the structural similarity of that local window. The average of the structural similarities of all windows in the entire image is taken as the SSIM value for the entire image. The SSIM value ranges from 0 to 1, with values ​​closer to 1 indicating more complete preservation of image structural features. Finally, a weighted fusion is performed using a core protection zone mask to analyze the P values ​​within the core protection zone. The PSNR and SSIM values ​​are weighted at 0.7, while the background area is weighted at 0.3. The final distortion metric is calculated by weighting these values. A higher PSNR value and an SSIM value closer to 1 indicate lower distortion and better compressed image quality. The weighting of PSNR and SSIM values ​​is based on the statistical analysis of 500 sets of satellite remote sensing images of reservoirs. The core protection area accounts for only 15% to 25% of the total map area, but contains 100% of the key features for safety monitoring, contributing 92% to the effectiveness of image quality. The background area accounts for 75% to 85% of the total map area, has no key monitoring features, and contributes only 8% to the effectiveness of image quality. The weights of 0.7 and 0.3 strictly match the actual contribution ratio of the core area and the background area, ensuring that the distortion metric is highly consistent with the actual needs of reservoir safety monitoring.

[0063] The preset distortion threshold was calibrated using 500 sets of satellite remote sensing images of reservoirs. It was set as follows: PSNR ≥ 35dB for the weighted core protection area and SSIM ≥ 0.85 for the entire image. This threshold ensures that key monitoring features such as dam cracks and water level fluctuations are not significantly distorted, meeting the accuracy requirements of remote sensing monitoring in the water conservancy industry. If the calculated distortion metric meets the preset threshold requirements, i.e., PSNR ≥ 35dB for the core protection area and SSIM ≥ 0.85 for the entire image, the hybrid compressed image is deemed to have passed the quality check. A fixed-length 16-bit binary verification pass identifier is then generated. This identifier contains the image acquisition timestamp, core area proportion, compression ratio, cyclic redundancy check, and four other key pieces of information, which are used by the receiving end to quickly verify the integrity and validity of the image.

[0064] The verification code is written into the auxiliary information segment at the end of the hybrid compressed image stream, and the core protection zone mask and key encoding / decoding parameters are written simultaneously to ensure that the receiving end can directly extract relevant information to complete decoding and reconstruction. Finally, the image is encapsulated and generated to conform to the satellite communication transmission protocol. If the distortion metric exceeds the preset threshold, the compressed image quality is deemed substandard, triggering a recompression process. The stream is sent back to step S16, where the hybrid compressed image is regenerated by adjusting the rate-distortion optimization truncation point, widening the quantization step size in the background area, and prioritizing the integrity of the core area stream. The quality verification in this step is then performed again until the distortion meets the threshold requirements.

[0065] In step S18, multi-layer resolution decoding is performed on the final transmitted image at the receiving end, and the hybrid compressed image is restored and reconstructed according to the core protection zone mask to obtain a complete image of the reservoir dam body and water level safety monitoring.

[0066] It should be noted that when reading the header and trailer auxiliary information segments of the final transmitted image bitstream, the checksum at the trailer is first verified to confirm that the image has passed quality verification before transmission and that there has been no bitstream loss or tampering during transmission. If the checksum verification fails, a retransmission request is triggered. After successful verification, the metadata in the bitstream header is completely parsed to extract key parameters that are completely consistent with those at the encoding end, including image size, band information, wavelet basis type, quantization step size table for each sub-band, wavelet decomposition level, core protection zone mask, coding block segmentation rules, and geographic coordinate information. Subsequently, the data is parsed according to the JPEG2000 standard. The accurate multi-layer resolution decoding rule performs layer-by-layer decoding on the embedded code block data in the final transmitted image. First, the code stream is subjected to adaptive arithmetic entropy decoding to restore the wavelet coefficients after quantization at the encoding end. Then, based on the region division mark at the head of the code stream, the decoded code stream data is separated into a core compressed data stream and a background compressed data stream. The core compressed data stream corresponds to the core area data of dam safety monitoring generated by the encoding end S15, and the background compressed data stream corresponds to the non-core background area data generated by the encoding end S16. At the same time, the parsed core protection area mask and key encoding and decoding parameters are synchronously cached.

[0067] The core protection zone mask is mapped to the pixel coordinate system of the decoding end to accurately locate the spatial region corresponding to the core compressed data stream. The pixel coordinate marked as 1 in the mask is the core protection zone that needs to be restored without loss. It matches the mask coordinates generated by the encoding end S14 perfectly without spatial offset. Then, the core compressed data stream is subjected to inverse quantization processing corresponding to the encoding end. The low quantization step size of the core region in the encoding end S15 is adopted, namely, the low frequency sub-band step size 3, the horizontal and vertical high frequency sub-band step size 4, and the diagonal high frequency sub-band step size 6. The wavelet coefficients obtained by decoding are inversely quantized point by point to completely restore the wavelet coefficient values ​​before encoding without loss of quantization accuracy. Then, a three-level inverse discrete wavelet transform is performed on the inversely quantized wavelet coefficients. The same Daubechies9 / 7 biorthogonal wavelet basis as the encoding end is adopted to complete the inverse transform layer by layer from the high frequency sub-band to the low frequency sub-band. Finally, a lossless restored pixel block is generated that is completely consistent with the pixel grayscale values ​​of the original image at the encoding end. This pixel block completely preserves the key details of reservoir safety monitoring such as dam cracks, water level fluctuation traces, and abnormal dam leakage.

[0068] The separated background compressed data stream is subjected to inverse quantization and inverse discrete wavelet transform corresponding to the encoder. The high quantization step size of the background region in encoder S16 is adopted, namely, low frequency sub-band step size 32, horizontal and vertical high frequency sub-band step size 48, and diagonal high frequency sub-band step size 64. After inverse quantization, a fourth-level inverse discrete wavelet transform is performed to restore the low frequency contour pixel matrix of the background region. This matrix retains the overall spatial structure of the background region, but high frequency details are lost due to high quantization step size compression, and compression block effect also exists.

[0069] For different land cover types in the background area, a context-adaptive interpolation algorithm is used for detailed reconstruction. A 3×3 analysis window is constructed with each pixel to be interpolated as the center. The gray values, gradient directions, and texture features of the eight known neighboring pixels in the window are extracted as context information. Linear interpolation is used to smooth the transition in uniform water areas and eliminate block effects. Bicubic interpolation is used to preserve basic texture features in textured areas such as vegetation and mountains around the reservoir. An edge-preserving interpolation algorithm is used in non-core bank slope areas around the dam to avoid edge blurring caused by interpolation. Pixel gray value constraints are set simultaneously during the interpolation process to ensure that the gray value difference between the reconstructed background pixel value and the gray value at the boundary of the adjacent core area does not exceed eight gray levels. Finally, a background area reconstruction detail matrix with smooth details, no obvious block effects, and natural transition with the boundary of the core area is generated.

[0070] Specifically, by selecting 500 samples of reservoir background areas, three types of areas—uniform water bodies, vegetation, and non-core bank slopes—were labeled to form a standard dataset. The mean and gray-level variance distribution of near-infrared reflectance of each pixel within the labeled areas were statistically analyzed. Based on the statistical distribution results, the discrimination threshold for uniform water body areas was set as a mean near-infrared reflectance of less than 0.10 and a gray-level variance of less than 5; the discrimination threshold for vegetation areas was a red band to near-infrared reflectance ratio between 1.5 and 3.0 and a gray-level variance of more than 10; and the non-core bank slope areas were the remaining cases.

[0071] Extract the boundary coordinates of the lossless restored pixel block. These coordinates perfectly match the outer contour of the core protection area mask. At the same time, verify the pixel continuity of the background region reconstruction detail matrix at the boundary. If the gray level difference between adjacent pixels on both sides of the boundary is less than 8 preset gray levels, the boundary is considered to be continuous and valid. The boundary coordinates are directly extracted for subsequent stitching. If the boundary continuity does not meet the requirements, perform secondary interpolation optimization on the pixels at the boundary of the background region until the continuity constraint is met.

[0072] Subsequently, based on the boundary coordinates, the lossless restored core pixel blocks are precisely overlaid and stitched to the corresponding positions of the reconstructed detail matrix in the background area. During the stitching process, a feathering blending process with a width of 4 to 6 pixels is applied at the boundary between the core area and the background area, so that the pixel grayscale values ​​on both sides of the boundary change linearly, completely eliminating the stitching traces. The feathering blending uses a linear transition band with a width of 5 pixels, with the boundary of the core area as the 0 point, extending 2 pixels into the core area and 3 pixels into the background area to form a transition band. Finally, the stitched full-frame image is bound and encapsulated with the geographic coordinate information parsed from the bitstream to generate a complete image of the reservoir dam and water level safety monitoring with the same size, coordinate system, and band range as the original clear image output by S11. The core monitoring area of ​​this image has no distortion in detail and the background area has visual continuity.

[0073] The feathering fusion width is determined by testing transition bands with widths of 3, 4, 5, 6, and 7 pixels respectively. The continuity of the fused image boundary is subjectively scored, and the gray-level gradient changes of pixels on both sides of the boundary are calculated. When the width is 5 pixels, the gradient change at the boundary is the smoothest, and the subjective score is the highest. Therefore, 5 pixels is selected as the optimal fusion width.

[0074] In summary, this invention discloses an image compression and transmission method for satellite image transmission monitoring stations in reservoirs, which can solve the image distortion problems caused by traditional solutions that focus on compression while neglecting details, lack of verification during transmission, and distortion during decoding.

[0075] Reference Figure 2 The second embodiment of the present invention provides an image compression and transmission system for a satellite image transmission monitoring station for a reservoir, comprising: The data acquisition module is used to acquire multispectral raw image data of the reservoir area, remove systematic errors from the multispectral raw image data, and obtain a cleared image. The data extraction module is used to extract the edge intensity and texture gradient of the multispectral band from the sharpened image. If the edge intensity and the texture gradient both exceed the preset intensity threshold and gradient threshold, the corresponding pixel points are extracted and connected component morphological closing operation is performed to obtain the boundary of the core target region. The pixel expansion module is used to statistically analyze the multispectral pixel distribution histogram within the boundary of the core target region and expand it outward to adjacent pixels to obtain the expanded core region. The mask generation module is used to generate a multispectral saliency map based on the extended core region, traverse the pixel values ​​in the multispectral saliency map to determine the relevant areas for reservoir dam safety monitoring, and obtain the core protection zone mask. The data decomposition module is used to perform block-based three-level discrete wavelet transform decomposition on the extended core region to generate a sub-band coefficient matrix, and to perform low-quantization step-size encoding on the sub-band coefficient matrix according to the core protection zone mask to obtain core compressed data. The image compression module is used to perform a fourth-level discrete wavelet transform on the background region in the sharpened image that does not belong to the extended core region, and perform high-quantization step-size encoding to generate background compressed data. The core compressed data and the background compressed data are then fused to obtain a hybrid compressed image. The data calculation module is used to extract embedded code block data from the hybrid compressed image, calculate a distortion metric value based on the embedded code block data, and if the distortion metric value is lower than a preset distortion threshold, then the hybrid compressed image is confirmed as the final transmission image. The image decoding module is used to perform multi-layer resolution decoding on the final transmitted image at the receiving end, and to restore and reconstruct the hybrid compressed image according to the core protection zone mask to obtain a complete image of the reservoir dam and water level safety monitoring.

[0076] It should be noted that the satellite image transmission and monitoring station image compression and transmission system for reservoirs provided in this embodiment of the invention is used to execute all the process steps of the satellite image transmission and monitoring station image compression and transmission method for reservoirs in the above embodiment. The working principles and beneficial effects of the two are one-to-one, so they will not be described again.

[0077] It should be noted that the system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the system embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0078] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. A method for image compression and transmission from a satellite image transmission monitoring station for a reservoir, characterized in that, include: Acquire multispectral raw image data of the reservoir area, remove systematic errors from the multispectral raw image data, and obtain a sharpened image; The edge intensity and texture gradient of the multispectral band are extracted from the sharpened image. If the edge intensity and texture gradient both exceed the preset intensity threshold and the preset gradient threshold, the corresponding pixel is extracted and connected component morphological closing operation is performed to obtain the boundary of the core target region. The extended core region is obtained by statistically analyzing the multispectral pixel distribution histogram within the boundary of the core target region and extending it outward to adjacent pixels. A multispectral saliency map is generated based on the extended core region. The pixel values ​​in the multispectral saliency map are traversed to determine the relevant areas for reservoir dam safety monitoring, and the core protection zone mask is obtained. The extended core region is decomposed into a block-based three-level discrete wavelet transform to generate a sub-band coefficient matrix. The sub-band coefficient matrix is ​​then encoded with a low quantization step size according to the core protection zone mask to obtain core compressed data. A fourth-level discrete wavelet transform is performed on the background regions in the sharpened image that do not belong to the extended core region, and high-quantization step-size encoding is performed to generate background compressed data. The core compressed data and the background compressed data are then fused to obtain a hybrid compressed image. Embedded code block data is extracted from the hybrid compressed image, and a distortion metric is calculated based on the embedded code block data. If the distortion metric is lower than a preset distortion threshold, the hybrid compressed image is confirmed as the final transmission image. At the receiving end, multi-layer resolution decoding is performed on the final transmitted image, and the hybrid compressed image is restored and reconstructed according to the core protection zone mask to obtain a complete image of the reservoir dam and water level safety monitoring.

2. The image compression and transmission method for satellite image transmission monitoring stations used in reservoirs according to claim 1, characterized in that, The process of acquiring multispectral raw image data of the reservoir area, removing systematic errors from the multispectral raw image data, and obtaining a sharpened image includes: Extract the sensor response grayscale values ​​from the original multispectral image data, and perform a linear transformation based on the sensor response grayscale values ​​to obtain an absolute radiance image; Calculate the pixel offset of the absolute radiance image. If the pixel offset is greater than a preset offset threshold, resample the absolute radiance image to obtain a resampled pixel array. Orthorectification is performed based on the resampled pixel array to obtain a sharpened image.

3. The image compression and transmission method for satellite image transmission monitoring stations used in reservoirs according to claim 1, characterized in that, The process involves extracting multispectral band edge intensity and texture gradient from the sharpened image. If both the edge intensity and texture gradient exceed a preset intensity threshold and gradient threshold, the corresponding pixels are extracted and connected component morphological closing operations are performed to obtain the boundary of the core target region, including: Multispectral bands are extracted from the sharpened image, a Sobel template is applied to the multispectral bands, the gradient magnitudes in the horizontal and vertical directions are calculated, and the edge intensity of the multispectral bands is synthesized. Local windows in the sharpened image are located based on the edge intensity, and the local windows are processed by a preset gray-level co-occurrence matrix to obtain the texture gradient; If the edge intensity and the texture gradient both exceed the preset intensity threshold and gradient threshold, then the corresponding pixel points are extracted to obtain a set of high-frequency abrupt pixel points; Perform connected component morphological closing operation on the set of high-frequency abruptly changing pixels to obtain the boundary of the core target region.

4. The image compression and transmission method for a satellite image transmission monitoring station for a reservoir according to claim 1, characterized in that, The step of statistically analyzing the multispectral pixel distribution histogram within the boundary of the core target region and extending it outward to adjacent pixels to obtain the extended core region includes: Extract the multispectral pixel values ​​within the boundary of the core target region, and construct a multispectral pixel distribution histogram based on the multispectral pixel values; Statistical feature parameters are calculated based on the pixel distribution histogram, and the similarity between the statistical feature parameters and adjacent pixels is quantified to obtain a similarity metric. The extended range is obtained by performing region growing based on the similarity metric. The extended area is then filled with holes to obtain the extended core region.

5. The image compression and transmission method for a satellite image transmission monitoring station for a reservoir according to claim 1, characterized in that, The process of generating a multispectral saliency map based on the extended core region, traversing the pixel values ​​in the multispectral saliency map to determine the relevant areas for reservoir dam safety monitoring, and obtaining the core protection zone mask includes: Multidimensional local contrast features are calculated based on the multispectral reflectance data within the extended core region, and a multispectral saliency map is generated based on the multidimensional local contrast features. The pixel values ​​of the multispectral saliency map are traversed. If a pixel value is greater than a preset adaptive threshold, it is marked as a high saliency score pixel. Connectivity component labeling is performed on the highly significant fractional pixels to obtain the relevant area for reservoir dam safety monitoring; A binary image matrix is ​​generated based on the relevant area of ​​the reservoir dam safety monitoring. The elements in the binary image matrix are assigned mask identifiers to obtain the core protection area mask.

6. The image compression and transmission method for a satellite image transmission monitoring station for a reservoir according to claim 1, characterized in that, The extended core region is decomposed into a block-based three-level discrete wavelet transform to generate a sub-band coefficient matrix. The sub-band coefficient matrix is ​​then encoded with a low-quantization step size based on the core protection zone mask to obtain core compressed data, including: Based on the core protection zone mask, a block-based three-level discrete wavelet transform decomposition is performed on the extended core region to generate a sub-band coefficient matrix; The core protection zone mask is mapped to a pre-constructed transform domain coordinate system to obtain the core coefficient index set corresponding to the sub-band coefficient matrix; If the coordinates of the sub-band coefficient matrix belong to the core coefficient index set, then the coefficients corresponding to the coordinates of the sub-band coefficient matrix are quantized to obtain high-precision quantized coefficients. The high-precision quantization coefficients are encoded to generate a compressed bitstream, and the compressed bitstream is encapsulated to obtain core compressed data.

7. The image compression and transmission method for a satellite image transmission monitoring station for a reservoir according to claim 1, characterized in that, The process involves performing a fourth-level discrete wavelet transform on the background regions in the sharpened image that do not belong to the extended core region, followed by high-quantization step-size encoding to generate background compressed data. The core compressed data is then fused with the background compressed data to obtain a hybrid compressed image, comprising: A fourth-level discrete wavelet transform is performed on the background regions in the sharpened image that do not belong to the extended core region, and high-quantization step-size encoding is performed to generate background compressed data. Calculate the rate-distortion slope based on the background compressed data, sort the coding blocks corresponding to the rate-distortion slope from high to low to construct a sorting table, and obtain a global sorting table. Traverse the global sorting table to find the optimal cutoff point and obtain the optimal cutoff point position; The core compressed data and the background compressed data are truncated and spliced ​​together according to the optimal truncation point to obtain a hybrid compressed image.

8. The image compression and transmission method for a satellite image transmission monitoring station for a reservoir according to claim 1, characterized in that, The step of extracting embedded code block data from the hybrid compressed image, calculating a distortion metric based on the embedded code block data, and confirming the hybrid compressed image as the final transmission image if the distortion metric is lower than a preset distortion threshold includes: Embedded code block data is extracted from the hybrid compressed image, and the embedded code block data is inversely transformed to generate a locally reconstructed pixel matrix; The difference feature matrix is ​​obtained by performing a difference calculation between the locally reconstructed pixel matrix and the original multispectral image data; The distortion degree is calculated based on the difference feature matrix to obtain the distortion metric value; If the distortion metric is lower than a preset distortion threshold, a verification pass identifier is generated and written into the hybrid compressed image to obtain the final transmitted image.

9. An image compression and transmission system for a satellite image transmission monitoring station in a reservoir, characterized in that, include: The data acquisition module is used to acquire multispectral raw image data of the reservoir area, remove systematic errors from the multispectral raw image data, and obtain a cleared image. The data extraction module is used to extract the edge intensity and texture gradient of the multispectral band from the sharpened image. If the edge intensity and the texture gradient both exceed the preset intensity threshold and gradient threshold, the corresponding pixel points are extracted and connected component morphological closing operation is performed to obtain the boundary of the core target region. The pixel expansion module is used to statistically analyze the multispectral pixel distribution histogram within the boundary of the core target region and expand it outward to adjacent pixels to obtain the expanded core region. The mask generation module is used to generate a multispectral saliency map based on the extended core region, traverse the pixel values ​​in the multispectral saliency map to determine the relevant areas for reservoir dam safety monitoring, and obtain the core protection zone mask. The data decomposition module is used to perform block-based three-level discrete wavelet transform decomposition on the extended core region to generate a sub-band coefficient matrix, and to perform low-quantization step-size encoding on the sub-band coefficient matrix according to the core protection zone mask to obtain core compressed data. The image compression module is used to perform a fourth-level discrete wavelet transform on the background region in the sharpened image that does not belong to the extended core region, and perform high-quantization step-size encoding to generate background compressed data. The core compressed data and the background compressed data are then fused to obtain a hybrid compressed image. The data calculation module is used to extract embedded code block data from the hybrid compressed image, calculate a distortion metric value based on the embedded code block data, and if the distortion metric value is lower than a preset distortion threshold, then the hybrid compressed image is confirmed as the final transmission image. The image decoding module is used to perform multi-layer resolution decoding on the final transmitted image at the receiving end, and to restore and reconstruct the hybrid compressed image according to the core protection zone mask to obtain a complete image of the reservoir dam and water level safety monitoring.

Images