Data compression and reconstruction method and system based on survey image saliency region

Abstract

The embodiment of the invention provides a data compression and reconstruction method and system based on a survey image salient region, and belongs to the technical field of survey image data processing. The data compression and reconstruction method comprises the following steps: acquiring a to-be-compressed survey image; performing feature extraction on the survey image, and preprocessing the features; obtaining potential feature representation according to the preprocessed features; obtaining a salient region according to the potential feature representation, and carrying out feature coding to obtain a compressed code stream; recovering and acquiring potential feature representation according to the compressed code stream; obtaining initial image data according to the recovered potential feature representation; according to the method, the saliency information in the survey image can be effectively reserved in a mode of obtaining the saliency region according to the potential feature representation, the information loss in the data compression process is effectively reduced, and the compression effect is improved; meanwhile, image enhancement processing is carried out during image reconstruction, so that the quality of image reconstruction is further improved.

Description

Technical Field

[0001] The present invention relates to the technical field of survey image data processing, and particularly to a data compression and reconstruction method and system based on the salient regions of survey images. Background Art

[0002] With the advancement of the digital age, the survey field is gradually undergoing a transformation towards high-precision, real-time, and intelligent. In application scenarios such as remote sensing images, geological exploration, and environmental monitoring, the large-scale survey data generated daily is growing exponentially. However, these data usually have characteristics such as high resolution, large amount of information, and complex features, posing severe challenges to storage, transmission, and subsequent analysis. Traditional survey image data processing technologies are difficult to meet the requirements of efficient transmission and accurate reconstruction under limited computing resources and network bandwidth, and are prone to problems such as data loss, low compression ratio, or poor reconstruction quality.

[0003] To address the bottleneck problems of storage and transmission in survey data, in recent years, image compression methods based on deep learning have gradually attracted attention. This method realizes efficient compression and reconstruction by automatically learning the data distribution characteristics. However, existing methods have problems such as loss of important information during the compression process, poor compression effect, and poor reconstruction quality.

[0004] The inventors of the present application found during the implementation of the present invention that the above solutions of the prior art have defects of poor compression effect and poor reconstruction quality. Summary of the Invention

[0005] The objective of the embodiments of the present invention is to provide a data compression and reconstruction method and system based on the salient regions of survey images, which have the functions of good compression effect and good reconstruction quality.

[0006] To achieve the above objective, the embodiments of the present invention provide a data compression and reconstruction method based on the salient regions of survey images, including:

[0007] Obtain a survey image to be compressed;

[0008] Extract features from the survey image and preprocess the features;

[0009] Obtain a latent feature representation according to the preprocessed features;

[0010] Obtain the salient regions according to the latent feature representation and perform feature encoding to obtain a compressed bitstream;

[0011] Restore and obtain the latent feature representation according to the compressed bitstream;

[0012] Obtain preliminary image data according to the restored latent feature representation;

[0013] Perform enhancement processing on the preliminary image data to obtain a reconstructed image.

[0014] Optionally, feature extraction of the survey image includes: extracting features from the survey image using a convolution operation.

[0015] Optionally, preprocessing of the features includes: rearranging and optimizing the features using a shuffle operation.

[0016] Optionally, obtaining a saliency region and performing feature encoding according to the latent feature representation to obtain a compressed bitstream includes:

[0017] Input the latent feature representation into a saliency region determination module to obtain corresponding saliency weights;

[0018] Obtain a dynamically adjusted quantization step according to the saliency weights of the latent feature representation;

[0019] Perform block quantization according to the dynamically adjusted quantization step of the latent feature representation and obtain the quantized feature values;

[0020] Obtain a compressed bitstream according to the quantized feature values.

[0021] Optionally, obtaining a dynamically adjusted quantization step according to the saliency weights of the latent feature representation includes:

[0022] Obtain a dynamically adjusted quantization step according to formula (1),

[0023]

[0024] where Δ(x) is the dynamically adjusted quantization step of the latent feature representation, Δ base is the basic quantization step, and A s (x) is the saliency weight of the latent feature representation.

[0025] Optionally, performing block quantization according to the dynamically adjusted quantization step of the latent feature representation and obtaining the quantized feature values includes:

[0026] Obtain the quantized feature values according to formula (2),

[0027]

[0028] where, is the quantized feature value, Round() is the rounding operation, max and min are respectively the maximum and minimum values of the latent feature values, and z is the input latent feature value.

[0029] Optionally, obtaining the compressed bitstream according to the quantized eigenvalue includes:

[0030] Obtaining the compressed bitstream according to formula (3),

[0031]

[0032] where R(x) is the compressed bitstream and λ is the scale factor, is the compression operation on the quantized eigenvalue .

[0033] Optionally, obtaining the preliminary image data according to the restored latent feature representation includes: inputting the restored latent feature representation into a decoder to obtain the preliminary image data.

[0034] Optionally, performing enhancement processing on the preliminary image data to obtain the reconstructed image includes:

[0035] Obtaining the feature map of the preliminary image data;

[0036] Performing dimensionality increase on the feature map by using a convolution operation to obtain the dimensionality-increased data;

[0037] Inputting the dimensionality-increased data into a residual block and obtaining the final output, where the residual block includes a main path and a skip connection;

[0038] Inputting the final output into a convolutional layer for dimensionality reduction to obtain the reconstructed image.

[0039] On the other hand, the present invention also provides a data compression and reconstruction system based on the salient region of a survey image, including:

[0040] An encoding module, configured to receive a survey image to be compressed and output the latent feature representation of the survey image;

[0041] A salient region dynamic compression module, with an input end connected to the encoding module, configured to perform dynamic quantization and encoding on the latent feature representation and output a compressed bitstream;

[0042] An inverse entropy decoding and inverse quantization module, with an input end connected to the output end of the salient region dynamic compression module, configured to process the compressed bitstream to restore the latent feature representation;

[0043] A decoding module, with an input end connected to the output end of the inverse entropy decoding and inverse quantization module, configured to output the reconstructed image;

[0044] A controller, connected to the encoding module, the salient region dynamic compression module, the inverse entropy decoding and inverse quantization module, and the decoding module, configured to execute any one of the above data compression and reconstruction methods.

[0045] With the above technical solution, a data compression and reconstruction method and system based on the salient region of a survey image provided by the present invention preprocesses the survey image by feature extraction to obtain a corresponding latent feature representation. The salient region can be determined according to the latent feature representation and feature encoding is performed, and then a compressed bitstream is obtained to achieve the compression of the survey image. Further, the compressed bitstream is restored to the latent feature representation and converted into preliminary image data, and then the preliminary image data is enhanced to achieve the reconstruction of the survey image. By obtaining the salient region according to the latent feature representation, the salient information in the survey image can be effectively retained, the information loss in the data compression process is effectively reduced, and the compression effect is improved. At the same time, image enhancement is performed during image reconstruction, further improving the quality of image reconstruction.

[0046] Other features and advantages of the embodiments of the present invention will be described in detail in the following specific implementation part. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] The drawings are used to provide a further understanding of the embodiments of the present invention, and constitute a part of the specification, and are used to explain the embodiments of the present invention together with the following specific implementation manners, but do not constitute a limitation to the embodiments of the present invention. In the drawings:

[0048] Figure 1 is a flowchart of a data compression and reconstruction method based on the salient region of a survey image according to an embodiment of the present invention;

[0049] Figure 2 is a flowchart of compressing the latent feature representation in a data compression and reconstruction method based on the salient region of a survey image according to an embodiment of the present invention;

[0050] Figure 3 is a flowchart of image reconstruction in a data compression and reconstruction method based on the salient region of a survey image according to an embodiment of the present invention;

[0051] Figure 4 is a schematic diagram of a data compression and reconstruction system based on the salient region of a survey image according to an embodiment of the present invention;

[0052] Figure 5 is a schematic diagram of obtaining the latent feature representation in a data compression and reconstruction method based on the salient region of a survey image according to an embodiment of the present invention;

[0053] Figure 6 is a schematic diagram of performing image quality enhancement in a data compression and reconstruction method based on the salient region of a survey image according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0054] The following will detail the specific implementation manners of the embodiments of the present invention in conjunction with the accompanying drawings. It should be understood that the specific implementation manners described herein are only for explaining and illustrating the embodiments of the present invention, and are not used to limit the embodiments of the present invention.

[0055] It should be noted that the acquisition, transmission, storage, use, processing, etc. of data in the technical solution of this application all comply with the relevant regulations of laws and regulations. In the embodiments of this application, some existing solutions in the industry such as certain software, components, models, etc. may be mentioned. They should be regarded as exemplary, and their purpose is only to illustrate the feasibility in the implementation of the technical solution of this application, but it does not mean that the applicant has already or necessarily used this solution.

[0056] Figure 1 is a flowchart of a data compression and reconstruction method based on the salient region of a survey image according to an embodiment of the present invention. Specifically, in Figure 1 this method, the data compression and reconstruction method may include:

[0057] In step S1, a survey image to be compressed is acquired. Among them, the sources of the survey image may include survey equipment, drone shooting, and other data sources.

[0058] In step S2, features of the survey image are extracted and the features are preprocessed. As Figure 5 shown, after the survey image to be compressed is acquired, its features are extracted, and after the extraction is completed, preprocessing is performed to improve the subsequent compression performance. Specifically, the feature extraction of the survey image includes extracting features from the survey image using a convolution operation. The preprocessing of the features includes reordering and optimizing the features using a shuffle operation. Specifically, by introducing a shuffle attention module, features can be efficiently extracted, and the feature weight distribution can be optimized through the shuffle operation and the channel attention mechanism, thereby enhancing the feature expression ability of the salient region and reducing the interference of redundant information.

[0059] In step S3, a latent feature representation is obtained based on the preprocessed features. Among them, for the latent feature representation, it may include obtaining it by weighting and optimizing the features in combination with an attention mechanism, and the purpose of obtaining the latent feature representation is to be used as the input for the subsequent compression processing step. Specifically, obtaining the corresponding latent feature representation according to the survey image corresponds to Figure 4 the image transformation network in, that is, the encoding module. Specifically, obtaining the latent feature representation may include, as shown in formula (4),

[0060] z = FeatureExtracgtion(X h,w,c ) (4)

[0061] Among them, z is the latent feature representation extracted from the input data, with the shape of a feature tensor the same as that of the input data, and X h,w,c is the input data.

[0062] In step S4, the saliency region is obtained based on the latent feature representation and feature encoding is performed to obtain a compressed bitstream. Among them, the saliency region and the non-saliency region are determined according to the latent feature representation, and differential quantization processing is performed on the saliency region and the non-saliency region. Compression is performed on the latent feature representation after differential quantization processing to form a compressed bitstream, that is, feature encoding of the saliency region dynamic compression module is performed on the latent feature representation.

[0063] In step S5, the latent feature representation is restored and obtained according to the compressed bitstream. Among them, when performing image reconstruction on the compressed data (compressed bitstream), it is necessary to first restore the compressed bitstream into the latent feature representation. Specifically, inverse entropy decoding and inverse quantization processing can be performed through the inverse entropy decoding and inverse quantization module to restore the compressed bitstream into the latent feature representation. Specifically, the compressed bitstream / bitstream can be restored to the quantized feature through inverse entropy decoding as shown in formula (6),

[0064]

[0065] Among them, is the quantized feature value, and R(x) is the compressed bitstream.

[0066] The inverse quantization process is performed on the quantized feature value according to formula (7) to restore the latent feature representation,

[0067]

[0068] Among them, Δ(x) is the quantization step size.

[0069] In step S6, the preliminary image data is obtained according to the restored latent feature representation. Among them, after restoring the latent feature representation, it can be input into the decoder / decode module (image inverse transformation network) to restore the preliminary image data.

[0070] In step S7, the preliminary image data is enhanced to obtain the reconstructed image. Among them, in order to further improve the effect of image reconstruction, the preliminary image data also needs to be processed for image enhancement to obtain the final reconstructed image. Specifically, by embedding a quality enhancement module in the decoding module, the restored latent feature representation can be converted into the reconstructed image, that is, corresponding to Figure 4 the image inverse transformation network in.

[0071] In steps S1 to S7, when it is necessary to compress the survey image, first obtain the survey image to be compressed, and perform feature extraction and preprocessing operations on the survey image. According to the features after preprocessing, obtain the corresponding latent feature representation, and determine the significant region and the non-significant region based on the latent feature representation. Then, perform differential processing on the significant region and the non-significant region, that is, perform differential feature encoding to obtain the compressed bitstream, and thus complete the compression of the survey image to be compressed. When it is necessary to compress data for image reconstruction, restore the compressed bitstream into the latent feature representation, and obtain the corresponding preliminary image data based on the latent feature representation. Finally, perform enhancement processing on the preliminary image data to obtain a high-quality reconstructed image.

[0072] Traditional survey image data uses a deep learning-based image compression method, but this method has problems such as loss of important information during compression, poor compression effect, and poor reconstruction quality. In this embodiment of the present invention, the method of using latent features to obtain significant regions and perform differential feature encoding can significantly reduce information loss during data compression, improve the compression effect, and maintain the detail fidelity and global consistency of data under high compression ratio conditions. At the same time, image enhancement processing is performed during image reconstruction, improving the quality and effect of image reconstruction, and showing excellent reconstruction ability under high compression ratio and complex field scenarios. The present invention achieves a new balance between compression efficiency and reconstruction quality through the organic combination of feature extraction, dynamic compression of significant regions, and quality enhancement.

[0073] In this embodiment of the present invention, in order to perform differential feature encoding / compression on the latent feature representation and achieve effective compression of the significant region and the non-significant region, it is also necessary to determine the significant region based on the latent feature representation. Specifically, the compression steps can be as Figure 2 shown. Specifically, in Figure 2 , this data compression and reconstruction method may further include:

[0074] In step S40, input the latent feature representation into the significant region determination module to obtain the corresponding significant weight. Among them, the significant region determination module may include a deep learning network model, which can be trained using historical data. Specifically, the significant determination module can generate the significant weight A s (x) in combination with the attention mechanism. The region with a high significant weight A s (x) is determined as the significant region and is preferentially encoded with high quality. Specifically, the acquisition of the significant weight may include, as shown in formula (5),

[0075] A s (x) = Attention(z), (5)

[0076] Among them, Attention(z) is an attention mechanism module that weights regions of the feature map based on attention.

[0077] In step S41, a dynamically adjusted quantization step size is obtained according to the significance weight of the latent feature representation. Among them, for obtaining the dynamic quantization step size, it can be dynamically adjusted according to the significance weight. Specifically, it can include obtaining the dynamically adjusted quantization step size according to formula (1).

[0078]

[0079] Among them, Δ(x) is the dynamically adjusted quantization step size of the latent feature representation, and Δ base is the base quantization step size, and A s (x) is the significance weight of the latent feature representation. Specifically, the base quantization step size Δ base is a globally set constant, and the significance weight A s (x) is used to measure the significance degree of a certain region. The larger the value, the more significant the region is and the more key information it contains. Specifically, for a significant region A s (x) being large, Δ(x) decreases, the quantization precision increases, and more details are retained; for a non-significant region A s (x) being small, Δ(x) increases, the quantization step size is larger, and storage resources are saved. Traditional quantization generally uses a fixed step size. This formula combines the significance weight with the idea of dynamic quantization, and can reduce information loss in non-significant regions while preferentially allocating resources to significant regions.

[0080] In step S42, partition quantization is performed according to the dynamically adjusted quantization step size of the latent feature representation, and the quantized feature value is obtained. Among them, after obtaining the dynamically adjusted quantization step size of the latent feature representation, partition quantization can be performed on the latent feature representation. Specifically, it can include obtaining the quantized feature value according to formula (2).

[0081]

[0082] Among them, is the quantized feature value, Round() is the rounding operation, z is the input latent feature value, and max and min are respectively the maximum and minimum values of the latent feature value. Specifically, this formula realizes two key steps of the quantization operation. One is to scale the latent feature value (latent feature representation) z to a standard range [0, 1], and the other is to perform quantization and rounding operations on the scaled feature value according to the step size Δ(x).

[0083] In step S43, a compressed bitstream is obtained based on the quantized eigenvalue. After obtaining the significance weight and the quantized eigenvalue, differential encoding compression can be performed on the significant region and the non-significant region. Specifically, dynamic entropy encoding can be used, including obtaining the compressed bitstream according to formula (3).

[0084]

[0085] where R(x) is the compressed bitstream, that is, the compressed code stream. λ is a scaling factor used to adjust the influence of the significance weight. For the compression operation of the quantized eigenvalue is used to improve the efficiency of entropy encoding. Specifically, this dynamic entropy encoding can dynamically allocate bit resources based on the significance weight, allocate more bits to the significant region to ensure the reconstruction quality, allocate fewer bits to the non-significant region, and the generated compressed code stream can be used for storage or transmission. By combining the significance weight with the logarithmic function design, it is ensured that more bit resources are preferentially allocated to the significant region, while reducing the storage requirements of the non-significant region, thereby effectively avoiding / reducing the occurrence of important information loss and improving the compression effect and compression efficiency.

[0086] In steps S40 to S43, the obtained latent feature representation is input into the significant region determination module to obtain the corresponding significance weight, and the significant region can be determined according to this significance weight. Then, according to this significance weight, the dynamically adjusted quantization step size of the latent feature representation is determined, and the latent feature representation is partitioned and quantized according to this quantization step size to obtain the quantized eigenvalue. Finally, differential compression / encoding is performed according to the quantized eigenvalue to obtain the compressed code stream to achieve the compression of the survey image data. By dynamically adjusting the quantization step size based on the significance weight, fine-grained compression processing is performed on the significant region, and a differential quantization strategy is adopted for the non-significant region, which reduces data redundancy while ensuring the precise retention of significant information. Specifically, the entropy encoding and inverse entropy decoding mechanism adopted by the present invention combines the dynamically adjusted quantization step size, making the compressed bitstream achieve high efficiency and stability during storage and transmission.

[0087] In this embodiment of the present invention, during the process of image reconstruction, the latent feature representation can be restored through the compressed code stream, and the preliminary image data can be restored through the latent feature representation. To improve the effect of the reconstructed image, the preliminary image data also needs to be enhanced, and the specific processing steps can be as Figure 3 and Figure 6 shown. Specifically, in Figure 3 and Figure 6 , this data compression and reconstruction method may further include:

[0088] In step S70, a feature map of the preliminary image data is obtained.

[0089] In step S71, a convolution operation is adopted to perform dimensionality increase on the feature map to obtain dimensionality-increased data.

[0090] In step S72, the dimensionality-increased data is input into a residual block, and a final output is obtained, where the residual block includes a main path and a skip connection. Specifically, the residual block includes a main path and a skip connection. The main path is composed of a convolutional layer and an activation function. The skip connection realizes the direct transfer of the input to the output of the convolutional block by adding the input and the output of the main path. That is, the final output is the residual function of the output of the main path and the skip connection of the input.

[0091] In step S73, the final output is input into a convolutional layer for dimensionality reduction to obtain a reconstructed image.

[0092] In steps S70 to S73, first, a feature map of the preliminary image data is obtained, and then a convolution operation is adopted to perform dimensionality increase on the feature map to obtain corresponding dimensionality-increased data. The dimensionality-increased data is input into a residual block, and the residual function of the output of the main path in the residual block and the skip connection of the input are used as the final output. Finally, the final output is input into a convolutional layer for dimensionality reduction, and thus, the enhancement processing of the preliminary image data can be realized to obtain a high-quality reconstructed image. Specifically, through the residual learning path and the skip connection technology, the capture of image details can be enhanced, thereby enhancing the reconstruction quality of the image. The edge details and high-frequency information performance of the reconstructed image are improved through the quality enhancement module, and finally, high-compression-rate and high-fidelity data transmission and reconstruction are realized.

[0093] On the other hand, the present invention also provides a data compression and reconstruction system based on the salient region of a survey image. Specifically, as Figure 4 shown, the data compression and reconstruction system may include an encoding module, a salient region dynamic compression module, an entropy decoding and inverse quantization module, and a decoding module.

[0094] The encoding module is configured to receive a survey image to be compressed and output a latent feature representation of the survey image. The input end of the salient region dynamic compression module is connected to the encoding module and is configured to perform dynamic quantization and encoding on the latent feature representation and output a compressed bitstream. The input end of the entropy decoding and inverse quantization module is connected to the output end of the salient region dynamic compression module and is configured to process the compressed bitstream to restore the latent feature representation. The input end of the decoding module is connected to the output end of the entropy decoding and inverse quantization module and is configured to output a reconstructed image. The controller is connected to the encoding module, the salient region dynamic compression module, the entropy decoding and inverse quantization module, and the decoding module and is configured to execute any one of the above data compression and reconstruction methods.

[0095] The encoding module, namely the encoder, consists of multiple layers of convolution, normalization (GDN), and an attention module (MS-SA), and is responsible for extracting potential feature representations. Specifically, Figure 4 Conv(N, 3) / 2 in means using a convolution kernel of 3×3 with a stride of 2 to extract image features. GDN is used to normalize features to improve the network's expressive ability. The decoder restores and reconstructs the image through inverse normalization (iGDN), transposed convolution, etc., and at the same time embeds a quality enhancement module to improve the image quality.

[0096] Through the above technical solution, a data compression and reconstruction method based on the salient regions of survey images provided by the present invention preprocesses by extracting features from the survey images to obtain corresponding potential feature representations. The salient regions can be determined based on the potential feature representations and feature encoding is performed, and then a compressed bitstream is obtained to achieve the compression of the survey images; further, the compressed bitstream is restored to a potential feature representation, converted into preliminary image data, and then the preliminary image data is enhanced to achieve the reconstruction of the survey images. By obtaining the salient regions according to the potential feature representations, the salient information in the survey images can be effectively retained, the information loss in the data compression process is effectively reduced, and the compression effect is improved; at the same time, image enhancement processing is performed during image reconstruction, further improving the quality of image reconstruction.

[0097] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0098] The present application is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or block in the flowcharts and / or block diagrams, and the combination of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, such that the instructions executed by the processor of the computer or other programmable data processing devices generate means for implementing the functions specified in Figure 1 one process or multiple processes and / or blocks Figure 1 one block or multiple blocks.

[0099] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to operate in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including an instruction means that implements the functions specified in one or more of the processes Figure 1 one or more of the processes and / or blocks Figure 1 specified in one or more of the blocks or blocks.

[0100] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process, so that the instructions executed on the computer or other programmable apparatus provide steps for implementing the functions specified in one or more of the processes Figure 1 one or more of the processes and / or blocks Figure 1 specified in one or more of the blocks or blocks.

[0101] In a typical configuration, a computing device includes one or more processors (CPUs), an input / output interface, a network interface, and memory.

[0102] The memory may include non-permanent memory in the form of computer-readable media, random access memory (RAM) and / or non-volatile memory such as read-only memory (ROM) or flash memory (flash RAM). Memory is an example of computer-readable media.

[0103] Computer-readable media includes both permanent and non-permanent, removable and non-removable media implemented by any method or technology for storing information. The information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, compact disc read-only memory (CD-ROM), digital versatile discs (DVD) or other optical storage, magnetic cassettes, magnetic tape disk storage or other magnetic storage devices, or any other non-transitory media that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transitory media such as modulated data signals and carrier waves.

[0104] It should also be noted that the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, such that a process, method, commodity or device comprising a series of elements not only includes those elements but also includes other elements not expressly listed, or elements inherent to such process, method, commodity or device. Without further limitation, an element defined by the statement "comprising an..." does not exclude the presence of additional identical elements in the process, method, commodity or device comprising the element.

[0105] The above are only embodiments of the present application and are not intended to limit the present application. For those skilled in the art, various changes and modifications can be made to the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included within the scope of the claims of the present application.

Claims

1. A data compression and reconstruction method based on the salient regions of survey images, characterized in that, Including: Obtain a survey image to be compressed; Extract features from the survey image and preprocess the features; Obtain a potential feature representation according to the preprocessed features; Obtain a saliency region according to the potential feature representation and perform feature encoding to obtain a compressed bitstream; Restore and obtain a potential feature representation according to the compressed bitstream; Obtain preliminary image data according to the restored potential feature representation; Perform enhancement processing on the preliminary image data to obtain a reconstructed image.

2. The data compression and reconstruction method according to claim 1, wherein Extracting features from the survey image includes: extracting features from the survey image by using a convolution operation.

3. The data compression and reconstruction method according to claim 1, wherein Preprocessing the features includes: rearranging and optimizing the features by using a shuffle operation.

4. The data compression and reconstruction method according to claim 1, characterized in that Obtaining a saliency region according to the potential feature representation and performing feature encoding to obtain a compressed bitstream includes: Input the potential feature representation into a saliency region determination module to obtain corresponding saliency weights; Obtain a dynamically adjusted quantization step according to the saliency weights of the potential feature representation; Perform partition quantization according to the dynamically adjusted quantization step of the potential feature representation and obtain quantized feature values; Obtain a compressed bitstream according to the quantized feature values.

5. The data compression and reconstruction method according to claim 4, characterized in that, Obtaining a dynamically adjusted quantization step according to the saliency weights of the potential feature representation includes: Obtain a dynamically adjusted quantization step according to formula (1), where Δ(x) is the quantization step for dynamic adjustment of the potential feature representation, and Δ base is the basic quantization step, and A s (x) is the significance weight of the potential feature representation.

6. The data compression and reconstruction method according to claim 5, wherein Performing partition quantization according to the dynamically adjusted quantization step of the potential feature representation and obtaining quantized feature values includes: Obtain quantized feature values according to formula (2), Among them, is the quantized eigenvalue, Round() is the rounding operation, max and min are the maximum and minimum values of the potential eigenvalue respectively, and z is the input potential eigenvalue.

7. The data compression and reconstruction method according to claim 6, wherein, Obtaining a compressed bitstream according to the quantized feature values includes: Obtain a compressed bitstream according to formula (3), Among them, R(x) is the compressed bitstream, and λ is the scaling factor. It is the compression operation on the quantized eigenvalue .

8. The data compression and reconstruction method according to claim 1, wherein Obtaining preliminary image data according to the restored potential feature representation includes: inputting the restored potential feature representation into a decoder to obtain preliminary image data.

9. The data compression and reconstruction method according to claim 1, wherein Performing enhancement processing on the preliminary image data to obtain a reconstructed image includes: Obtain a feature map of the preliminary image data; Perform dimension elevation on the feature map by using a convolution operation to obtain dimension-elevated data; Input the dimension-elevated data into a residual block and obtain a final output, where the residual block includes a main path and a skip connection; Input the final output into a convolutional layer for dimension reduction to obtain a reconstructed image.

10. A data compression and reconstruction system based on the salient regions of survey images, characterized in that, Including: An encoding module, configured to receive a survey image to be compressed and output a potential feature representation of the survey image; A saliency region dynamic compression module, with an input end connected to the encoding module, configured to perform dynamic quantization and encoding on the potential feature representation and output a compressed bitstream; An anti-entropy decoding and anti-quantization module, with an input end connected to the output end of the saliency region dynamic compression module, configured to process the compressed bitstream to restore a potential feature representation; A decoding module, with an input end connected to the output end of the anti-entropy decoding and anti-quantization module, configured to output a reconstructed image; A controller, connected to the encoding module, the saliency region dynamic compression module, the anti-entropy decoding and anti-quantization module, and the decoding module, configured to execute the data compression and reconstruction method according to any one of claims 1-9.

Images