A lossless and near-lossless compression method for GNSS reflectometry DDM bright areas

Abstract

The present application belongs to the technical field of spaceborne data compression and encoding, and particularly relates to a GNSS reflection DDM bright area lossless and background near-lossless compression method, which makes the bright area information lossless, the background area realizes near-lossless reconstruction under the controllable distortion constraint of a preset error boundary, and adopts a low-complexity encoding mode suitable for the limited spaceborne resource condition to realize stable decoding and reconstruction, and is used for reducing the data amount pressure of DDM data download under the condition of limited spaceborne storage and space-ground link bandwidth.

Description

Technical Field

[0001] This invention belongs to the field of spaceborne data compression and coding technology, specifically relating to a lossless compression method for bright areas and near-lossless compression of background in GNSS reflection DDM. Background Technology

[0002] Global Navigation Satellite System Reflection Remote Sensing (GNSS-R) utilizes the echoes of navigation satellite signals scattered across the Earth's surface for observation. The Delay-Doppler Map (DDM) is one of the crucial fundamental observations of GNSS-R. It is obtained by correlating the received scattered signal with the local signal model over a series of delay and Doppler frequency grids, and is used to characterize the energy distribution of the scattering contribution area near the mirror point in the delay-Doppler domain. Taking publicly available algorithmic theoretical documents from missions such as CYGNSS as examples, the DDM is used as an important input for subsequent inversions such as sea surface wind speed. Its typical shape exhibits a horseshoe-like distribution and is related to the scattering physics and geometry.

[0003] Because GNSS-R payloads need to continuously track and process signals from multiple navigation satellites, Data Metrics and Analysis (DDM) is typically generated in a "multi-channel, multi-mirror point, continuous time series" manner, with each observation being a two-dimensional matrix of data. This results in significant data volume pressure under conditions of limited onboard storage and satellite-to-ground link bandwidth. Existing engineering missions have adopted a mode of compressing the DDM before downloading it. Taking the theoretical basis document of the CYGNSS wind speed product algorithm published by NOAA as an example, it points out that a compressed DDM can be generated for each mirror point, with a typical size of 11 Doppler grids × 17 delay grids, and the data around the mirror point is extracted for subsequent NBRCS estimation and product generation. At the same time, the theoretical document of the CYGNSS Level-1A DDM calibration algorithm also publicly explains the DDM processing link.

[0004] In addition to compressed DDM, the publicly available data product description also points out that: conventional DDM typically samples only within a limited delay and Doppler range centered on the mirror point, and the bit resolution of the scattered signal intensity is truncated by lossy data compression algorithms; in contrast, Full DDM samples over a wider delay and Doppler range and retains the full bit resolution, thus complementing scientific applications where conventional compressed DDM is insufficient in terms of size and bit width.

[0005] In addition to the compression methods specific to GNSS-R missions, the aerospace field has also developed a general data compression standard system. For example, CCSDS121 provides a recommended standard for lossless data compression, and CCSDS123 provides a recommended standard for low-complexity lossless and near-lossless multispectral image compression, which can be used for compressed transmission of spaceborne data.

[0006] However, for data with obvious physical structure characteristics, such as delay-Doppler maps, existing technologies still have the following shortcomings:

[0007] (1) The overall window limitation and bit resolution truncation of compressed DDM belong to "globally consistent processing": Public product specifications indicate that conventional DDM usually has limited sampling near the specular point, and the scattering intensity bit resolution is lossily compressed and truncated. This approach does not distinguish between the different accuracy requirements of bright areas near the specular point and background noise areas far away from the bright area, making it difficult to simultaneously achieve a balance between "higher fidelity requirements in bright areas" and "higher bit rate optimization space in background areas".

[0008] (2) DDM has horseshoe-shaped structural features, but existing methods do not provide a partitioning error control mechanism for this structure: the publicly available ATBD points out that DDM can exhibit a typical horseshoe-shaped distribution. Existing compression modes rely more on uniform window clipping, uniform bit width truncation, or quantization, and do not provide a mechanism to implement controllable errors and perform efficient encoding of the background area while keeping the bright area lossless.

[0009] (3) The applicable objects and data structures of the general compression standards are different: CCSDS 121 is for lossless compression encapsulation and encoding processes for general sequence data, while CCSDS 123 is for low-complexity lossless and near-lossless compression methods and bitstream formats for multi-hyperspectral and hyperspectral images. The above standards themselves do not provide dedicated processing procedures for the partitioning fidelity requirements and structural characteristics of the "mirror illumination area - background area" of GNSS-R DDM. Therefore, under the limited computing power of spaceborne systems, a low-complexity differentiation coding scheme for the characteristics of DDM is still needed to achieve stable decoding and reconstruction and controllable errors. Summary of the Invention

[0010] To solve the above technical problems, the present invention provides a method for lossless compression of the bright area and near-lossless compression of the background in GNSS reflection DDM, comprising:

[0011] S1. Read and parse the DDM two-dimensional matrix data of the target frame, and preprocess the DDM frame data to obtain the two-dimensional matrix data to be compressed.

[0012] S2. Perform peak location and estimate background noise level for the two-dimensional matrix data to be compressed. Generate candidate bright areas in the neighborhood of the peaks based on a preset threshold. Then, use the peaks as seeds to perform connected component extraction to obtain the mask of the bright area region ROI.

[0013] S3. Generate compressed bitstream header information, wherein the header information includes at least matrix size parameters, background region error bounds, quantization step size parameters, and block and encoding parameter constraint information;

[0014] S4. Write the ROI mask into the bitstream, and write the ROI region pixel values ​​into the bitstream in a lossless manner according to a predetermined scanning order;

[0015] S5. Perform two-dimensional prediction on background area pixels based on reconstructed neighborhood values ​​to obtain predicted values, calculate prediction residuals and quantize them according to preset error boundaries to obtain quantization residuals.

[0016] S6. Map the quantization residual to a non-negative integer sequence, group it according to a preset block length, compress each group of data using a low-complexity entropy encoding method, and write it into the bitstream.

[0017] S7. At the receiving end, parse the code stream header information and ROI mask, restore the lossless data of the ROI region, and decode, dequantize and reconstruct the background region data to obtain the reconstructed DDM.

[0018] The beneficial effects of this invention are:

[0019] This invention utilizes the characteristics of concentrated energy in bright areas and noise-dominated backgrounds in DDM (Digital Dispersion Metrics) for partitioning. Data in bright areas of the Region of Interest (ROI) is written losslessly to maintain the accuracy of key scattering information. In the background area, closed-loop prediction, residual quantization, and Rice low-complexity entropy coding are employed under error bounds to achieve controllable distortion and improved compression ratio, thereby reducing data downlink pressure. Compared with existing technologies that use overall pruning, uniform bit width truncation, or uniform quantization for DDM compression, this invention preserves key scientific information with higher fidelity, controllable distortion, and a clear upper bound on error. Simultaneously, the bitstream carries header fields and block-level parameters such as size, quantization step size, block length, and coding parameters, ensuring self-description, unique decodability, and improved decoding and reconstruction stability. Attached Figure Description

[0020] Figure 1 A compression flowchart provided for this invention;

[0021] Figure 2 This invention provides a flowchart for ROI bright area extraction.

[0022] Figure 3 A flowchart for background area compression provided by the present invention;

[0023] Figure 4 The 2D predictive scanning flowchart provided by this invention;

[0024] Figure 5 The flowchart of Rice encoding by length J blocks provided by this invention;

[0025] Figure 6 The bitstream format diagram provided by this invention;

[0026] Figure 7 The error distribution diagram after compression and decompression provided by this invention;

[0027] Figure 8 The linear relationship between compression ratio and error provided by this invention. Detailed Implementation

[0028] 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.

[0029] A method for lossless and near-lossless compression of bright areas and background in GNSS reflection DDM (Digital Direct Memory Access) includes the following steps: 1. Reading and parsing the DDM two-dimensional matrix data of the target frame, and preprocessing the DDM frame data to obtain the two-dimensional matrix data to be compressed; 2. Performing peak localization and estimating the background noise floor in the two-dimensional matrix data, generating candidate bright areas in the peak neighborhood based on a preset threshold, and then performing connected component extraction using the peak as a seed to obtain a mask of the bright area region (ROI); 3. Generating compressed bitstream header information, the header information including at least matrix size parameters, background area error bounds, quantization step size parameters, and block and coding parameter constraint information; 4. Applying the ROI mask... 5. Write the ROI region pixel values ​​into the bitstream in a lossless manner according to a predetermined scanning order; 6. Perform two-dimensional prediction on the background region pixels based on the reconstructed neighborhood values ​​to obtain the predicted values, calculate the prediction residuals and quantize them according to a preset error bound to obtain the quantization residuals; 7. Map the quantization residuals into a non-negative integer sequence, group them according to a preset block length, compress each group of data using a low-complexity entropy coding method and write them into the bitstream; 8. At the receiving end, parse the bitstream header information and ROI mask, recover the lossless data of the ROI region, and decode, dequantize and reconstruct the background region data to obtain the reconstructed DDM, wherein the background region reconstruction error is constrained by the preset error bound.

[0030] Preprocessing the DDM frame data yields the two-dimensional matrix data to be compressed: non-finite values ​​are replaced with zeros, and outliers that do not conform to physical meaning are truncated, resulting in the matrix to be compressed. This preprocessing is used to ensure the numerical stability of subsequent peak localization, threshold segmentation, and encoding processes.

[0031] Peak localization, noise floor estimation, thresholding to generate candidate bright regions, and connected component extraction to obtain the ROI mask: in the matrix to be compressed Calculate the global maximum value and its location This location is used as the seed point for the bright area. Samples are then extracted from several rows / columns along the matrix edge, and the median is taken as the seed point. This reflects the background noise level. Then, using... Define the peak neighborhood window at the center (half-width in the delay direction) Doppler direction half-width ), and apply a threshold within that window:

[0032]

[0033] Binarization is performed to obtain the binary image of the candidate bright area. ;in This is a preset threshold coefficient. Finally, in... Seed point Starting from the seed point, perform connected component extraction (e.g., 8-connectivity), identify the connected regions containing the seed point as bright areas of interest (ROIs), and obtain the ROI mask. .

[0034] In the matrix to be compressed Calculate the global maximum value and its location ,include:

[0035] Iterate through the cells in the matrix X to be compressed, maintaining the maximum value X among the currently traversed cells. max and its coordinates (r) p ,c p ), initialize X max =X(1,1), for each pixel X traversed t If X t > X t-1 Then update X max =X t And record the coordinates as (r) p ,c p After traversing the matrix, the global maximum value and its position are obtained.

[0036] In the peak neighborhood window Seed point Perform connected component extraction starting from the point of origin, including:

[0037] Within the peak neighborhood window BW, with (r p ,c p Starting from ), a flood search is performed, continuously adding unvisited cells within the 8-neighborhood (where edges or corners are considered adjacent) that satisfy binarization to 1, until no new adjacent foreground cells can be expanded.

[0038] Generate compressed stream header information: synchronization identifier and version number; matrix size. Quantization step size Block length Entropy coding parameter constraints and flags indicating whether adaptive parameter selection is enabled; and the number of ROIs and background pixels. , This is used for consistency verification. The "header field + block-level parameters" code stream organization method is consistent with the engineering principle of "block processing and carrying necessary identifiers for unique decoding" in spatial data compression standards.

[0039] The ROI mask is written into the bitstream, and the ROI region pixel values ​​are written into the bitstream in a lossless manner according to a predetermined scanning order (preferably scanning column by column, first scanning the first column from top to bottom, then the second column, and so on). The ROI mask is written point by point according to the predetermined scanning order. Then, the matrix is ​​traversed in the same scanning order; when At the same time, the corresponding ROI cell values ​​are written to the bitstream in a lossless manner. In this embodiment, the cell values ​​are mapped to unsigned integers and written with a fixed word length, thereby ensuring that the receiver can recover the ROI bright area values ​​point by point without loss. At the same time, a reconstruction buffer is maintained at the encoding end. The ROI location is filled with the written lossless value, which is used as the reconstructed neighborhood for subsequent background prediction.

[0040] Two-dimensional prediction is performed on background pixels based on reconstructed neighborhood values ​​to obtain predicted values. The prediction residual is calculated and quantized according to a preset error bound to obtain the quantization residual. All background pixels are processed in the same scanning order, and prediction and reconstruction are completed in a closed-loop manner. The encoding end maintains a reconstruction buffer. and based on The reconstructed neighborhood is used to calculate the predicted value. Design a predictor when The predicted value is 0 when... The predicted value is taken from the left neighbor. ;when The predicted value is taken from the upper neighbor. In other cases, the predicted value is the average of the upper and left neighbors, rounded off. Calculate the predicted residuals, where The integerized pixel values ​​are then quantized according to a preset error bound to obtain the quantization residuals.

[0041]

[0042] Subsequently Obtain the reconstructed residuals and update the reconstructed values. This ensures that the encoder and decoder achieve consistent reconstruction under a consistent prediction context, and makes the background region reconstruction error subject to error bounds. constraint.

[0043] Quantitative residual It is a signed integer, which can be positive, 0, or negative. The Rice encoding used subsequently is designed for non-negative integers. Therefore, it is necessary to reversibly map the "signed integer" to the "non-negative integer" before encoding. The quantization residual is mapped to a sequence of non-negative integers, divided into blocks, and compressed using low-complexity entropy encoding before being written into the bitstream. Convert signed integers into sequences of nonnegative integers using an invertible one-to-one mapping. ,in season , season This is to enable variable length entropy encoding. The sequence... According to the preset block length Grouping and determining entropy coding parameters for each block. The The value of is limited to and the block Write the bitstream; then process each block Rice variable-length encoding is performed and written into the bitstream. The encoding is decomposed through shift and bitwise AND operations. and And represented by unary codes ,by Bit binary representation This reduces implementation complexity while ensuring decodability.

[0044] The receiver parses the header information and ROI mask, decodes the background, and inverse-quantizes to reconstruct the DDM. The receiver then reads the bitstream header information to recover the original data. The parameters and scanning order are agreed upon, and then the ROI mask is read according to the scanning order. And based on the header field and A consistency check is performed on the mask count. Then, the lossless values ​​of the ROI are read in the scanning order and written into the reconstruction matrix. ROI location; for background location, parameters are read block by block. And perform Rice decoding point by point to obtain non-negative integers. Then perform the inverse symbol folding mapping to obtain And inverse quantization to obtain The receiver uses the same predictor as the encoder to compute based on the reconstructed neighborhood. and execute Complete the reconstruction of the entire frame to obtain the reconstructed DDM, where the reconstruction error of the background area is constrained by a preset error boundary.

[0045] This embodiment uses the delayed-Doppler image (DDM) generated by a spaceborne GNSS reflection remote sensing payload as the processing object, and follows the... Figure 1 The overall process shown involves performing "lossless bright areas and near-lossless background" partitioned compression on the DDM frame, and achieving stable decoding and reconstruction at the receiving end.

[0046] In this embodiment, the DDM two-dimensional matrix data of the target frame is first read and parsed, and the frame is preprocessed to improve the numerical stability of subsequent peak localization and encoding processes. The preprocessing method involves replacing non-finite values ​​with zeros and truncating outliers that do not conform to physical meaning, thereby obtaining the two-dimensional matrix to be compressed. Subsequently, ROI bright area extraction is performed on the matrix to be compressed, such as... Figure 2 As shown in the flowchart, the global maximum value and its location are calculated on the matrix, and this location is used as the seed point for the bright area. Samples are then extracted from several rows and columns at the edge of the matrix, and the median is used as the background noise floor to reflect the background noise level. Next, with the seed point as the center, the half-width in the delay direction is set to 25, and the half-width in the Doppler direction is set to 10, thus defining the peak neighborhood window. Within this window, a threshold coefficient β is preset to 0.45, thereby generating a binary map of candidate bright areas. Finally, 8-connected component extraction is performed on the candidate bright area binary map starting from the seed point. The connected regions containing the seed point are determined as the bright area ROI, thus obtaining the ROI mask.

[0047] After completing the lossless writing of the ROI, the background area is processed according to... Figure 3 The flowchart shown illustrates near-lossless compression. In this embodiment, background pixels are compressed according to... Figure 4 The scanning sequence is shown, and a closed-loop approach is used to ensure that the encoder and decoder are in a consistent prediction context. The encoder calculates the predicted value based on the reconstructed neighboring pixels in the reconstruction buffer, and calculates the prediction residual accordingly. Then, it quantizes the residual according to a preset error bound to obtain the quantization residual. Subsequently, the reconstruction residual is obtained from the quantization residual, and the reconstruction value is updated, thereby ensuring that the reconstruction error of the background area is constrained by the preset error bound. As a low-complexity predictor implementation, the rule of "predicting 0 if there is no neighbor" is adopted at the boundary position. When only the left neighbor or the top neighbor exists, the left neighbor or the top neighbor is taken as the predicted value, respectively. When both the left neighbor and the top neighbor exist, the average of the two is taken and rounded to balance the implementation complexity and prediction effect.

[0048] Then, an invertible Zigzag one-to-one sign folding mapping is performed on the signed quantization residuals to convert them into a sequence of non-negative integers for variable-length encoding; such as Figure 5 As shown, the sequence is grouped according to the block length J, and a Rice encoding parameter k is determined for each block, with the value of k limited to a preset range. At the same time, the k of the block is written into the bitstream. Finally, Rice variable-length encoding is performed on each non-negative integer in the block and written into the bitstream.

[0049] Then press Figure 6As shown, compressed bitstream header information is generated. The bitstream header information includes at least: synchronization identifier and version number, matrix size parameters, background region error bounds or corresponding quantization step size parameters, block length, and entropy coding parameter constraint information. By using a "header field plus block-level parameters" organization method, the bitstream becomes self-descriptive and ensures unique decodability, thereby improving the stability of decoding and reconstruction at the receiving end. Subsequently, the ROI region cell values ​​are written into the bitstream in a lossless manner.

[0050] After compression, the bitstream is decompressed and compared with the data before compression. The error distribution is as follows: Figure 7 As shown, the linear relationship between compression ratio and error is as follows: Figure 8 As shown in the accompanying drawings and the above analysis, this invention can stably control the background area reconstruction error within a preset error range while ensuring high fidelity of key bright area information. It also significantly reduces the data bit rate and improves the reliability of satellite-to-ground link downlink and decoding reconstruction, thereby achieving better compression efficiency and engineering usability.

[0051] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for lossless compression of bright areas and near-lossless compression of background in GNSS reflection DDM, characterized in that, include: S1. Read and parse the DDM two-dimensional matrix data of the target frame, and preprocess the DDM frame data to obtain the two-dimensional matrix data to be compressed. S2. Perform peak location and estimate background noise level for the two-dimensional matrix data to be compressed. Generate candidate bright areas in the neighborhood of the peaks based on a preset threshold. Then, use the peaks as seeds to perform connected component extraction to obtain the mask of the bright area region ROI. S3. Generate compressed bitstream header information, wherein the header information includes at least matrix size parameters, background region error bounds, quantization step size parameters, and block and encoding parameter constraint information; S4. Write the ROI mask into the bitstream, and write the ROI region pixel values ​​into the bitstream in a lossless manner according to a predetermined scanning order; S5. Perform two-dimensional prediction on background area pixels based on reconstructed neighborhood values ​​to obtain predicted values, calculate prediction residuals and quantize them according to preset error boundaries to obtain quantization residuals. S6. Map the quantization residual to a non-negative integer sequence, group it according to a preset block length, compress each group of data using a low-complexity entropy encoding method, and write it into the bitstream. S7. At the receiving end, parse the code stream header information and ROI mask, restore the lossless data of the ROI region, and decode, dequantize and reconstruct the background region data to obtain the reconstructed DDM.

2. The method for lossless and near-lossless compression of bright areas and background in GNSS reflection DDM according to claim 1, characterized in that, The DDM frame data is preprocessed to obtain two-dimensional matrix data to be compressed, including: Non-finite values ​​are replaced with zeros, and outliers that do not conform to physical meaning are truncated to obtain the matrix to be compressed. .

3. The method for lossless and near-lossless compression of bright areas and background in GNSS reflection DDM according to claim 1, characterized in that, Peak location and background noise floor estimation are performed on the compressed two-dimensional matrix data. Candidate bright regions are generated in the neighborhood of the peaks based on a preset threshold. Then, connected component extraction is performed using the peaks as seeds to obtain a mask for the bright region ROI, including: In the matrix to be compressed Calculate the global maximum value and its location This location is used as the seed point for the bright area. Samples are then extracted from several rows / columns along the matrix edge, and the median is taken as the seed point. , A parameter reflecting the level of background noise; Next by position Define the peak neighborhood window around the center, with a half-width of [value] in the delay direction. Doppler half-width is The candidate bright area binary image is obtained by binarizing the data within the window using a preset threshold. Where T is a preset threshold; Finally, in the peak neighborhood window Seed point Starting from the seed point, perform connected component extraction, and determine the connected regions containing the seed point as bright areas of interest (ROIs), thus obtaining the ROI mask. Where H is the number of rows in the delay-Doppler matrix and W is the number of columns in the delay-Doppler matrix; The preset threshold includes: Where T is a preset threshold. This is the preset threshold coefficient.

4. The method for lossless and near-lossless compression of bright areas and background in GNSS reflection DDM according to claim 1, characterized in that, Generate compressed stream header information, including: Synchronization identifier and version number, matrix size Quantization step size Block length Entropy coding parameter constraints and whether adaptive parameter selection is enabled, as well as the number of bright ROIs and background pixels. , For use in consistency verification; where, The number of bright area ROIs. The number of background pixels. Here, H is the ROI mask, H is the number of rows in the delay-Doppler matrix, and W is the number of columns in the delay-Doppler matrix.

5. The method for lossless and near-lossless compression of bright areas and background in GNSS reflection DDM according to claim 1, characterized in that, Writing the ROI mask into the bitstream, and writing the ROI region pixel values ​​into the bitstream in a lossless manner according to a predetermined scanning order, includes: Write the ROI mask point by point according to the predetermined scanning sequence. Then, the matrix is ​​traversed in the same scanning order, when At that time, the corresponding ROI pixel value is written to the bitstream in a lossless manner, where, Let the value of the ROI mask be at the r-th row and c-th column. During the writing process, cell values ​​are mapped to unsigned integers and written in fixed word lengths, ensuring that the receiver can recover the ROI bright area values ​​point by point without loss; at the same time, a reconstruction buffer is maintained at the encoding end. The ROI location is filled with the written lossless value, which is used as the reconstructed neighborhood for subsequent background prediction.

6. The method for lossless and near-lossless compression of bright areas and background in GNSS reflection DDM according to claim 1, characterized in that, Two-dimensional prediction is performed on background pixels based on reconstructed neighborhood values ​​to obtain predicted values. The prediction residual is calculated and quantized according to a preset error bound to obtain the quantization residual, including: All background pixels are processed according to a predetermined scanning order, and prediction and reconstruction are completed using a closed-loop method; the encoding end maintains a reconstruction buffer. and based on The reconstructed neighborhood is used to calculate the predicted value; Design a predictor when The predicted value is 0 when... The predicted value is taken from the left neighbor. ;when The predicted value is taken from the upper neighbor. In other cases, the predicted value is the average of the upper and left neighbors and then rounded; where r is the row index of the r-th row of the ROI mask and c is the column index of the c-th column of the ROI mask. according to Calculate the predicted residuals and quantize them according to the preset error limits to obtain the quantized residuals. ;in, The predicted values ​​generated by the predictor Here, e represents the integerized pixel value, e represents the prediction residual, and E represents the error bound. Subsequently Obtain the reconstructed residuals and update the reconstructed values. This ensures that the encoder and decoder achieve consistent reconstruction under a consistent prediction context, and makes the background region reconstruction error subject to error bounds. Constraints; among which, To reconstruct the residuals.

7. The method for lossless and near-lossless compression of bright areas and background in GNSS reflection DDM according to claim 1, characterized in that, The quantization residual is mapped to a sequence of non-negative integers, grouped according to a preset block length, and each group of data is compressed and written into the bitstream using a low-complexity entropy encoding method, including: Signed quantization residuals Convert signed integers into sequences of nonnegative integers using an invertible one-to-one mapping. ,in season , season This is to enable variable length entropy encoding; will sequence According to the preset block length Grouping and determining entropy coding parameters for each block. The The value of is limited to and the block Write to the bitstream; where, This is the lower limit value of the encoding parameter k. This represents the upper limit of the encoding parameter k; Then, each block was examined. Rice variable-length encoding is performed and written into the bitstream; the Rice variable-length encoding is decomposed through shift and bitwise AND operations. and And represented by unary codes ,by Bit binary representation This reduces implementation complexity while ensuring decodability; among which, The quotient generated by the Rice coding rules. This represents the remainder generated by Rice encoding.

8. The method for lossless and near-lossless compression of bright areas and background in GNSS reflection DDM according to claim 1, characterized in that, At the receiving end, the bitstream header information and ROI mask are parsed to recover the lossless data in the ROI region. The background region data is then decoded, dequantized, and reconstructed to obtain the reconstructed DDM, including: The receiving end reads the bitstream header information to recover the data. The parameters and scanning order are agreed upon, and then the ROI mask is read in the scanning order. And based on the header field and A consistency check is performed on the mask count; where H is the number of rows in the delay-Doppler matrix and W is the number of columns in the delay-Doppler matrix. For residual quantization step size, The segment length for segmented Rice encoding. This is the lower limit value of the encoding parameter k. This represents the upper limit of the encoding parameter k; Next, the lossless values ​​of the ROI are read in the scanning order and written into the reconstruction matrix. The location of the ROI; For the background position, parameters are read in blocks. And perform Rice decoding point by point to obtain non-negative integers. Then perform the inverse symbol folding mapping to obtain And inverse quantization to obtain ;in, For inverse quantization residuals, A sequence of non-negative integers obtained by mapping a sequence of signed integers. To quantify the residual, This is the quantization step size; The receiver uses the same predictor as the encoder to compute based on the reconstructed neighborhood. and execute The entire frame is reconstructed to obtain the reconstructed DDM, where the reconstruction error of the background area is constrained by a preset error boundary.

Images