Satellite remote sensing image data on-orbit compression method and system

Abstract

The invention provides a satellite remote sensing image data on-orbit compression method and system. The method comprises the following steps: S1, row vector extraction: reading data line by line from an image data source to form a row vector; s2, performing line compression: performing instant compression processing on the line vector to obtain compressed line data, including S21, wavelet decomposition: performing wavelet decomposition on the line vector by adopting one-dimensional discrete wavelet transform to obtain wavelet decomposition coefficients including a low-frequency coefficient and a high-frequency coefficient and the length of each coefficient; s22, determining a compression threshold: determining the compression threshold based on the high-frequency coefficient and a predetermined energy retention ratio; s23, coefficient filtering: filtering high-frequency coefficients based on wavelet decomposition coefficients, the length of each coefficient and a compression threshold; and S3, recombining and outputting compressed data: taking the filtered high-frequency coefficient and low-frequency coefficient as input, performing wavelet inverse transformation to obtain a compressed row vector, and recombining row by row to obtain a compressed remote sensing image.

Description

Technical Field

[0001] This invention relates to the fields of satellite on-orbit data processing and digital image compression technology, and in particular to a method and system for on-orbit compression of satellite remote sensing image data. Background Technology

[0002] On-orbit compression algorithms for satellite remote sensing images can effectively reduce on-board data redundancy and transmit remote sensing data back to the ground in a timely manner using limited satellite-to-ground bandwidth, providing high-time data support for disaster relief, situational awareness, and other needs. Most existing mature satellite remote sensing image compression algorithms process entire images (such as JPEG2000 and PNG) or local blocks (e.g., 8×8 blocks in JPEG) as the processing object. Existing algorithms first transform the entire image or local block image into a new encoding / decoding space using methods such as frequency domain transformation, pixel value predictive coding, and deep learning compression. Then, within this new encoding / decoding space, data is compressed according to specific rules to reduce the redundancy of the original image data, thereby achieving lossy / lossless remote sensing image compression. Specifically, existing remote sensing image compression algorithms can be divided into three main categories based on their technical principles: transform-based compression, prediction-based compression, and learning-based compression.

[0003] Transform-based compression transforms an image from the spatial domain to another domain (such as the frequency domain), concentrating energy and making it easier to distinguish and discard unimportant information. Representative techniques include Discrete Wavelet Transform (DWT), Embedded Zerotree Wavelets (EZW), and Discrete Cosine Transform (DCT). DWT decomposes the image into sub-bands of different frequencies, retaining low-frequency approximations while compressing high-frequency details. A typical DWT algorithm is JPEG2000. EZW utilizes the correlation of wavelet coefficients across different scales (zero-tree structure) to achieve an embedded bitstream (which can be truncated and decoded as needed). It is characterized by high efficiency and is often used for real-time compression on satellite. A typical DCT-based compression algorithm is JPEG. Its core principle is to divide the remote sensing image into 8×8 local blocks, then transform each block to the frequency domain, retaining low-frequency coefficients and discarding high-frequency coefficients. This type of compression method is computationally fast, but it produces significant block artifacts at high compression ratios.

[0004] Prediction-based methods do not directly compress pixel values, but rather compress prediction errors. These algorithms first cache several rows of image data, then use the spatial or spectral redundancy of the image to predict the current pixel using already encoded neighboring pixels, encoding only the difference. Representative algorithms include JPEG-LS, LOCO-I, and LOCO-3D. JPEG-LS / LOCO-I uses the values ​​of neighboring pixels of the current pixel for prediction, encoding only the prediction error to achieve lossless or near-lossless compression. This type of compression method has low complexity and high fidelity; it is suitable for surveying and mapping fields with extremely high image quality requirements. LOCO-3D, based on JPEG-LS, further utilizes inter-frame (inter-multispectral band or time-series image) correlations for prediction, making it suitable for compressing hyperspectral images or multi-frame image sequences, achieving higher compression ratios.

[0005] Learning-based compression uses deep neural networks to automatically learn the optimal compressed representation of an image. It is a data-driven, end-to-end approach. Representative methods include Variational Autoencoders (VAEs), Generative Adversarial Networks (GANs), and diffusion models. VAEs compress images into "latent variables" through an encoder, and then reconstruct them through a decoder. Combined with entropy coding, they achieve a balance between bitrate and distortion. VAE performance has surpassed traditional hand-designed encoders (such as JPEG), but images become blurry at extremely low bitrates. GANs introduce "perceptual quality" optimization on top of compression, enabling the decoder to generate visually more realistic and texture-rich images, producing visually acceptable results at extremely low bitrates (e.g., <0.1 bpp).

[0006] The aforementioned satellite remote sensing image compression methods all follow a "readout-caching-global / local compression" process. Transform-based and learning-based compression methods process the entire image, while prediction-based compression methods require caching local images before data compression. In other words, existing image compression algorithms need to cache global or local images line by line before performing data compression. However, with the emergence and on-orbit application of high-throughput remote sensing imaging technologies such as high spatiotemporal resolution and wide swaths, the amount of satellite remote sensing data acquired has increased dramatically. The limited cache space on the onboard platform has led to a problem of "insufficient caching and inadequate compression" in onboard remote sensing data compression algorithms. First, the limited space of the onboard data caching module makes it difficult to cache high-throughput remote sensing data in a timely manner, easily causing data loss, i.e., "insufficient caching." Second, existing on-orbit remote sensing data compression algorithms require caching global or local remote sensing images before data compression. The surge in remote sensing data volume consumes more computing resources, which in turn affects the data compression effect, leading to "inadequate compression." To address the caching and compression challenges posed by the high-throughput remote sensing data, it is necessary to develop new on-orbit compression methods for satellite remote sensing images, improve data compression speed, and alleviate the pressure of on-board data caching and compression.

[0007] It should be noted that the information disclosed in the background section above is only for understanding the background of this application, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0008] To overcome the shortcomings of existing technologies, this invention provides an on-orbit compression method and system for satellite remote sensing image data, which solves the technical problems of existing on-orbit compression methods for satellite remote sensing images, such as limited on-board cache resources, high computational latency, and difficulty in adapting to the real-time processing requirements of high-throughput remote sensing data, due to the need for global or block-based image caching.

[0009] The present invention adopts the following technical solution: Firstly, an on-orbit compression method for satellite remote sensing image data is provided, comprising the following steps: S1, Row Vector Extraction: Read data row by row from the image data source to form row vectors; S2. Row compression: Performing real-time compression on the row vector to obtain compressed row data, wherein the row compression includes: S21. Wavelet decomposition: The row vector is decomposed by one-dimensional discrete wavelet transform to obtain wavelet decomposition coefficients and the length of each coefficient. The wavelet decomposition coefficients include low-frequency coefficients and high-frequency coefficients. S22. Determine the compression threshold: Determine the compression threshold for screening high-frequency coefficients based on the high-frequency coefficients and the predetermined energy retention ratio; S23. Coefficient filtering: Based on the wavelet decomposition coefficients, the length of each coefficient, and the compression threshold, the high-frequency coefficients are filtered. S3. Compressed data reconstruction and output: Using the filtered high-frequency coefficients and low-frequency coefficients as input, perform inverse wavelet transform to obtain compressed row vectors, and then reconstruct the compressed row vectors row by row to obtain compressed remote sensing images.

[0010] Further, in step S1, the row vector is a signal-level row vector or an image row vector; wherein: the signal-level row vector is a digitized signal value directly obtained from the detector readout circuit after analog-to-digital conversion, or the signal-level row vector is extracted row by row from a zero-level image; the image row vector is extracted row by row from a mature remote sensing image, which is obtained by preprocessing a zero-level image.

[0011] Furthermore, the mature remote sensing image is obtained by sequentially performing radiometric correction, atmospheric correction, image denoising, and image enhancement on a zero-level image.

[0012] Furthermore, in step S21, the one-dimensional discrete wavelet transform uses the Daubechies 4 wavelet function, with a decomposition level of 4.

[0013] Further, determining the compression threshold in step S22 includes: S221, extracting all high-frequency coefficients from the wavelet decomposition coefficients; S222, calculating the absolute value of each high-frequency coefficient and sorting them in descending order; calculating the sum of squares of all high-frequency coefficients as the total energy; S223, starting from the high-frequency coefficient with the largest absolute value, cyclically accumulating the squares of the sorted high-frequency coefficients to calculate the current energy, and when the current energy reaches a preset energy threshold, using the number of currently accumulated high-frequency coefficients as the threshold index; wherein, the preset energy threshold is the product of the predetermined energy retention ratio and the total energy; S224. Determine the compression threshold based on whether the threshold index exceeds the coefficient length. If it does not exceed the length, take the index corresponding to the sorted high-frequency coefficient as the compression threshold; otherwise, the compression threshold is 0.

[0014] Furthermore, the predetermined energy retention ratio is 95%.

[0015] Further, the coefficient filtering in step S23 includes: S231, using the wavelet decomposition coefficients, the length of each coefficient, and the compression threshold as input, first retaining the wavelet decomposition coefficients, determining the length of low-frequency coefficients, and not processing low-frequency coefficients; S232, iterating through each high-frequency coefficient, calculating its position in the coefficient vector, and extracting it; S233, processing the high-frequency coefficients according to the soft thresholding method: setting high-frequency coefficients with absolute values ​​lower than the compression threshold to zero, and retaining the original values ​​of high-frequency coefficients with absolute values ​​higher than the compression threshold.

[0016] In a second aspect, an on-orbit compression system for satellite remote sensing image data is provided, comprising: a processor; and a memory storing a computer program that can run on the processor, wherein the processor executes the computer program to implement the on-orbit compression method for satellite remote sensing image data as described in the first aspect.

[0017] Thirdly, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the on-orbit compression method for satellite remote sensing image data as described in the first aspect.

[0018] The beneficial effects of this invention include: The on-orbit compression method for satellite remote sensing image data provided by this invention, by reading line by line and compressing row vectors in real time, skips the necessary global or block image caching steps in traditional methods. This drastically reduces the amount of data processed per cycle from the number of pixels in the entire image (e.g., 1024x1024) to the number of pixels in a single row (e.g., 1024), greatly reducing the instantaneous demand on on-board cache space and fundamentally alleviating the problem of "insufficient cache space." Simultaneously, the significantly reduced amount of data processed per cycle significantly lowers the computational complexity and resource consumption of compression algorithms (e.g., wavelet transform), allowing the compression process to keep pace with the data readout rate, significantly improving the real-time performance and efficiency of compression processing, and solving the problem of "insufficient compression." Ultimately, this achieves the beneficial effects of improving the real-time performance of on-orbit compression and the timeliness of downlink for high-throughput satellite remote sensing data.

[0019] Furthermore, by defining two data sources—signal-level row vectors and image row vectors—this method can flexibly adapt to the outputs of different stages in the satellite data processing link, enhancing its versatility and practicality.

[0020] Furthermore, by employing the Daubechies 4 wavelet function and limiting the number of decomposition layers to 4, a good balance was achieved between computational complexity and compression effect, ensuring the feasibility and effectiveness of the method.

[0021] Furthermore, by using an adaptive threshold determination method based on the energy retention ratio, the main information can be intelligently preserved and redundant details can be filtered out, thus achieving a controllable balance between compression ratio and image quality.

[0022] Furthermore, by employing a soft thresholding method, detailed information such as image edges can be better preserved, resulting in a reconstructed image with superior subjective quality compared to a hard threshold. Attached Figure Description

[0023] Figure 1 This is a flowchart illustrating the on-orbit compression method for satellite remote sensing image data in an embodiment of the present invention. Figure 2 This is the original, uncompressed image in the embodiments of the present invention; Figure 3 This is the reconstructed image obtained after compression in the embodiments of the present invention; Figure 4 This is a diagram showing the difference between the images before and after compression in an embodiment of the present invention; Figure 5 This is a distribution diagram of the compression ratio of each row of the image in an embodiment of the present invention. Detailed Implementation

[0024] The embodiments of the present invention will be described in detail below. It should be emphasized that the following description is merely exemplary and not intended to limit the scope of the present invention or its application. Unless otherwise specified, the embodiments and features in the embodiments of this application can be combined with each other.

[0025] The core of this invention lies in changing the traditional "read-cache-compress" process and proposing a new approach of "line-by-line read-instant compression-reassembly." Its basic principle is that during the pipeline process of data reading from the detector or from the memory, each line of data (row vector) is immediately compressed based on one-dimensional discrete wavelet transform. After processing, it is then reassembled and transmitted. This invention decomposes the massive two-dimensional image compression problem into multiple one-dimensional row vector compression problems, greatly reducing the data size and cache requirements for a single processing operation. It is particularly suitable for on-orbit real-time compression scenarios of high-throughput remote sensing data.

[0026] See Figure 1 One embodiment of the present invention provides an on-orbit compression method for satellite remote sensing image data, comprising the following steps: S1. Row Vector Extraction: Read data row by row from the image data source to form row vectors. .

[0027] After a satellite optical remote sensing detector completes imaging and analog-to-digital conversion (A / D conversion), the detector's readout circuit outputs DN (Digital Number) values ​​(DN values ​​are the digitized signal values ​​output by the detector), i.e., signal-level row vectors, line by line. These signal-level row vectors are then typically cached line by line to form a zero-level image. The zero-level image undergoes various preprocessing steps (such as radiometric correction, atmospheric correction, image denoising, and image enhancement) to obtain a mature remote sensing image. According to existing technology, the mature remote sensing image is compressed before output, which is the traditional "imaging-caching-preprocessing-caching-compression" process. However, this invention does not require global or local image caching. It uses only signal-level row vectors or image row vectors (both of which are essentially 1*N dimensional vector data) as input for subsequent row compression, thereby skipping the image caching process and reducing onboard storage and computing resource consumption.

[0028] The signal-level row vector can be a digitized signal value directly obtained from the detector readout circuit after analog-to-digital conversion, or it can be extracted row by row from a zero-level image. The image row vector is extracted row by row from a mature remote sensing image.

[0029] S2. Row Compression: The row vectors are compressed in real-time to obtain compressed row data. Specifically, high-frequency coefficients are obtained through wavelet transform, and then low-information or highly redundant data is filtered out using adaptive thresholding, thereby achieving row vector compression. Ultimately, row-by-row compression of the satellite remote sensing image is achieved. The row compression includes: S21. Wavelet decomposition: The row vector is decomposed using a one-dimensional discrete wavelet transform to obtain the wavelet decomposition coefficients and the length of each coefficient. The wavelet decomposition coefficients include low-frequency coefficients and high-frequency coefficients.

[0030] Specifically, the row vectors are processed according to the following formula. Wavelet decomposition is performed to obtain high-frequency and low-frequency coefficients respectively: , , in, Low-frequency coefficients These are high-frequency coefficients. These are the low-frequency filter coefficients. These are the coefficients of the high-pass filter. n Let be the dimension of the row vector. In this embodiment, the basic wavelet function for row vector wavelet decomposition is set to db4 (Daubechies 4 wavelet function), and the number of decomposition levels is 4. After wavelet decomposition, the wavelet decomposition coefficients (including low-frequency and high-frequency coefficients at all levels) and the length of each coefficient are obtained.

[0031] S22. Determine the compression threshold: Determine the compression threshold for screening high-frequency coefficients based on the high-frequency coefficients and the predetermined energy retention ratio.

[0032] Specifically, after obtaining the wavelet coefficients of the row vectors, it is necessary to calculate the compression threshold to filter high-frequency coefficients. Determining the compression threshold includes: S221. Extract all high-frequency coefficients from the wavelet decomposition coefficients.

[0033] S222. Calculate the absolute value of each high-frequency coefficient and sort them in descending order; calculate the sum of squares of all high-frequency coefficients as the total energy.

[0034] S223. Starting with the high-frequency coefficient with the largest absolute value, the squares of the sorted high-frequency coefficients are accumulated cyclically to calculate the current energy. When the current energy reaches a preset energy threshold, the number of currently accumulated high-frequency coefficients is used as the threshold index. The preset energy threshold is the product of the predetermined energy retention ratio and the total energy. In this embodiment, the predetermined energy retention ratio is 95%. The energy retention ratio can be configured according to the actual task's requirements for image quality and compression rate. Increasing this ratio helps retain more image details but reduces the compression rate, and vice versa.

[0035] S224. Determine the compression threshold based on whether the threshold index exceeds the coefficient length. If it does not exceed the length, take the index corresponding to the sorted high-frequency coefficient as the compression threshold; otherwise, the compression threshold is 0.

[0036] S23. Coefficient Filtering: Based on the wavelet decomposition coefficients, the length of each coefficient, and the compression threshold, the high-frequency coefficients are filtered.

[0037] Coefficient filtering is a key step in line-by-line compression, used to filter data with low information content or high redundancy. Specifically, coefficient filtering includes: S231. Using the wavelet decomposition coefficients, the length of each coefficient, and the compression threshold as input, firstly, retain the wavelet decomposition coefficients, determine the length of the low-frequency coefficients, and do not process the low-frequency coefficients.

[0038] S232. Next, by iterating through each high-frequency coefficient, calculate its position in the coefficient vector and extract it.

[0039] S233. According to the soft thresholding method, the high-frequency coefficients are processed as follows: high-frequency coefficients with absolute values ​​lower than the compression threshold are set to zero, and high-frequency coefficients with absolute values ​​higher than the compression threshold retain their original values. Finally, the processed high-frequency coefficients are assigned back to the corresponding positions in the original coefficient vector, and the starting position of the next level of high-frequency coefficients is updated.

[0040] Among them, using a soft thresholding method can result in less loss of detail.

[0041] S3. Compressed data reconstruction and output: After coefficient filtering, the filtered wavelet coefficients (filtered high-frequency coefficients and unprocessed low-frequency coefficients) are retained. The filtered high-frequency coefficients and unprocessed low-frequency coefficients are used as input to perform inverse wavelet transform to obtain compressed row vectors. The compressed row vectors are then reconstructed row by row to obtain the compressed remote sensing image.

[0042] The compression method of this invention was fully simulated using computer simulation software. For a 1024*1024 resolution image, only 1024 pixels need to be processed at a time, while global wavelet transform compression requires processing 1024*1024 pixels. In comparison, the number of pixels processed per time is significantly reduced by this invention, making it more suitable for scenarios with limited onboard resources. Figures 2 to 5 As shown, Figure 2 This is an uncompressed, raw satellite infrared remote sensing image (original image), with a file size of 423KB. Figure 3 This is a remote sensing image (reconstructed image) obtained after line-by-line compression using the method described above in this invention. The file size is 303KB. Figure 4 This is a comparison image of the images before and after compression; the differences are only subtle in minor local areas. Figure 5 It represents the compression ratio of each row of the image, with the lowest compression ratio being 2 and the highest being nearly 10 times.

[0043] A specific embodiment of the present invention also provides an on-orbit compression system for satellite remote sensing image data, comprising: a processor; and a memory storing a computer program that can run on the processor, wherein the processor executes the computer program to implement the on-orbit compression method for satellite remote sensing image data.

[0044] A specific embodiment of the present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the on-orbit compression method for satellite remote sensing image data is implemented.

[0045] The row-vector-based satellite remote sensing data compression method in this invention skips the traditional "caching" step, reduces on-board cache resource consumption, and can effectively solve the problem of "insufficient caching and inadequate compression" in the high-throughput remote sensing data acquisition process, thereby achieving rapid high-throughput remote sensing data compression.

[0046] The above description provides a further detailed explanation of the present invention in conjunction with specific / preferred embodiments, and it should not be construed that the specific implementation of the present invention is limited to these descriptions. For those skilled in the art, various substitutions or modifications can be made to these described embodiments without departing from the concept of the present invention, and all such substitutions or modifications should be considered within the scope of protection of the present invention. In the description of this specification, the reference to terms such as "an embodiment," "some embodiments," "preferred embodiment," "example," "specific example," or "some examples," etc., indicates that the specific features, structures, materials, or characteristics described in connection with that embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described can be combined in any suitable manner in one or more embodiments or examples. Without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification and the features of different embodiments or examples. Although the embodiments of the present invention and their advantages have been described in detail, it should be understood that various changes, substitutions, and modifications can be made herein without departing from the scope of protection of the patent application.

Claims

1. A method for on-orbit compression of satellite remote sensing image data, characterized in that, Includes the following steps: S1, Row Vector Extraction: Read data row by row from the image data source to form row vectors; S2. Row compression: Performing real-time compression on the row vector to obtain compressed row data, wherein the row compression includes: S21. Wavelet decomposition: The row vector is decomposed by one-dimensional discrete wavelet transform to obtain wavelet decomposition coefficients and the length of each coefficient. The wavelet decomposition coefficients include low-frequency coefficients and high-frequency coefficients. S22. Determine the compression threshold: Determine the compression threshold for screening high-frequency coefficients based on the high-frequency coefficients and the predetermined energy retention ratio; S23. Coefficient filtering: Based on the wavelet decomposition coefficients, the length of each coefficient, and the compression threshold, the high-frequency coefficients are filtered. S3. Compressed data reconstruction and output: Using the filtered high-frequency coefficients and low-frequency coefficients as input, perform inverse wavelet transform to obtain compressed row vectors, and then reconstruct the compressed row vectors row by row to obtain compressed remote sensing images.

2. The on-orbit compression method for satellite remote sensing image data as described in claim 1, characterized in that, In step S1, the row vector is a signal-level row vector or an image row vector; wherein: The signal-level row vector is either a digital signal value directly obtained from the detector readout circuit after analog-to-digital conversion, or it is extracted row by row from the zero-level image. The image row vectors are extracted row by row from mature remote sensing images, which are obtained by preprocessing zero-level images.

3. The on-orbit compression method for satellite remote sensing image data as described in claim 2, characterized in that, The mature remote sensing image is obtained by sequentially applying radiometric correction, atmospheric correction, image denoising, and image enhancement to the zero-level image.

4. The on-orbit compression method for satellite remote sensing image data as described in claim 1, characterized in that, In step S21, the one-dimensional discrete wavelet transform uses the Daubechies 4 wavelet function, with a decomposition level of 4.

5. The on-orbit compression method for satellite remote sensing image data as described in claim 1, characterized in that, Determining the compression threshold in step S22 includes: S221. Extract all high-frequency coefficients from the wavelet decomposition coefficients; S222. Calculate the absolute value of each high-frequency coefficient and sort them in descending order; calculate the sum of squares of all high-frequency coefficients as the total energy. S223. Starting from the high-frequency coefficient with the largest absolute value, the squares of the high-frequency coefficients after sorting are accumulated in a loop to calculate the current energy. When the current energy reaches a preset energy threshold, the number of high-frequency coefficients accumulated at the moment is used as the threshold index; wherein, the preset energy threshold is the product of the predetermined energy retention ratio and the total energy. S224. Determine the compression threshold based on whether the threshold index exceeds the coefficient length. If it does not exceed the length, take the index corresponding to the sorted high-frequency coefficient as the compression threshold; otherwise, the compression threshold is 0.

6. The on-orbit compression method for satellite remote sensing image data as described in claim 5, characterized in that, The predetermined energy retention rate is 95%.

7. The on-orbit compression method for satellite remote sensing image data as described in claim 1, characterized in that, The coefficient filtering in step S23 includes: S231. Using the wavelet decomposition coefficients, the length of each coefficient, and the compression threshold as input, first retain the wavelet decomposition coefficients, determine the length of the low-frequency coefficients, and do not process the low-frequency coefficients. S232. Loop through each high-frequency coefficient, calculate its position in the coefficient vector, and extract it. S233. According to the soft threshold processing method, the high-frequency coefficients are processed: the high-frequency coefficients with absolute values ​​lower than the compression threshold are set to zero, and the high-frequency coefficients with absolute values ​​higher than the compression threshold are retained to their original values.

8. An on-orbit compression system for satellite remote sensing image data, characterized in that, include: processor; A memory having a computer program that can run on the processor, wherein the processor executes the computer program to implement the on-orbit compression method for satellite remote sensing image data as described in any one of claims 1 to 7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the on-orbit compression method for satellite remote sensing image data as described in any one of claims 1 to 7.

Images