Dynamic image transmission system based on deep learning

Abstract

The invention provides a dynamic image transmission system based on deep learning, and relates to the technical field of image transmission, and the system comprises the steps: collecting a to-be-transmitted dynamic image data flow through an image data collection unit; performing feature mining on the selected dynamic image through a scale feature extraction unit to obtain a scale feature vector, determining information singularity corresponding to the scale feature vector, and converting the scale feature vector into an information scale vector based on the information singularity; the image feature coding unit performs reconstruction according to the information scale vector to obtain a reconstructed image, and performs block coding on the reconstructed image by using a deep learning-based encoder to obtain image coding features; according to the technical scheme of the invention, the scale features in the dynamic image data can be effectively extracted and coded, the dynamic image data can be effectively extracted and coded, the dynamic image data can be effectively extracted and coded, and the dynamic image data can be effectively extracted and coded, so that the dynamic image data can be effectively extracted and coded, and the real-time data transmission can be realized. The transmission efficiency of dynamic image transmission is improved.

Description

Technical Field

[0001] The present application relates to the field of image transmission technology, and more specifically, to a dynamic image transmission system based on deep learning. Background Art

[0002] The dynamic image transmission system based on deep learning is an advanced solution that integrates deep learning technology and streaming media transmission protocols. It is designed to capture, encode and transmit dynamic image data in real time. The system captures dynamic images in real time through high-performance cameras or sensors, and then uses deep learning models (such as convolutional neural networks) to extract multi-scale features of the image. Multi-scale features can effectively capture important information in the image, and in the subsequent encoding process, the amount of data can be further reduced through appropriate compression algorithms, thereby improving transmission efficiency.

[0003] Deep learning models usually require a large amount of data for training. However, in dynamic scenes, there may be object occlusion, lighting changes, and motion blur, which affects the accuracy of feature extraction. In addition, the selection of feature encoding and compression algorithms is also crucial. Effective encoding technology not only needs to reduce the amount of data while maintaining image quality, but also takes into account computational complexity to meet the needs of real-time processing. Therefore, how to effectively extract and encode scale features in dynamic image data to improve the transmission efficiency of dynamic image transmission is a difficult problem faced by the industry. Summary of the invention

[0004] The present application provides a dynamic image transmission system based on deep learning, which can effectively extract and encode scale features in dynamic image data to improve the transmission efficiency of dynamic image transmission.

[0005] The present application provides a dynamic image transmission system based on deep learning, and the transmission system comprises:

[0006] An image data acquisition unit, which acquires a dynamic image data stream to be transmitted and selects a dynamic image from the dynamic image data stream;

[0007] A scale feature extraction unit performs feature mining on the selected dynamic image to obtain a scale feature vector of the selected dynamic image, further determines the information singularity corresponding to the scale feature vector, and converts the scale feature vector into an information scale vector of the selected dynamic image based on the information singularity;

[0008] An image feature encoding unit, reconstructing according to the information scale vector to obtain a reconstructed image corresponding to the selected dynamic image, and using a deep learning-based encoder to block-encode the reconstructed image to obtain an image coding feature corresponding to the selected dynamic image;

[0009] The coding feature transmission unit continues to select dynamic images in the dynamic image data stream, and then obtains the image coding features corresponding to the dynamic images at each moment in the dynamic image data stream, and performs real-time data transmission on the image coding features corresponding to the dynamic images at each moment.

[0010] In this embodiment, the dynamic image data stream to be transmitted is collected by a mobile camera.

[0011] In this embodiment, feature mining is performed on the selected dynamic image to obtain the scale feature vector of the selected dynamic image, specifically including:

[0012] Gray-scale the selected dynamic image to obtain a gray-scale image corresponding to the selected dynamic image;

[0013] constructing a Gaussian pyramid of a selected dynamic image based on the grayscale image;

[0014] Mining the image of each layer in the Gaussian pyramid to obtain the scale feature of the image of each layer in the Gaussian pyramid;

[0015] The scale feature vector of the selected dynamic image is determined through all the scale features.

[0016] In this embodiment, determining the information singularity corresponding to the scale feature vector specifically includes:

[0017] Constructing a covariance matrix of the scaled eigenvector;

[0018] Performing eigenvalue decomposition on the covariance matrix to obtain multiple eigenvalues ​​and eigenvectors corresponding to each eigenvalue;

[0019] Select the eigenvectors corresponding to the largest K eigenvalues, and then construct a feature matrix through all the selected eigenvectors;

[0020] The information singularity corresponding to the scale eigenvector is determined by the eigenmatrix.

[0021] In this embodiment, converting the scale feature vector into the information scale vector of the selected dynamic image based on the information singularity specifically includes:

[0022] constructing an orthogonal matrix of the scale eigenvectors;

[0023] The scale feature vector is transformed using the information singularity and the orthogonal matrix to obtain the information scale vector of the selected dynamic image.

[0024] In this embodiment, reconstructing according to the information scale vector to obtain a reconstructed image corresponding to the selected dynamic image specifically includes:

[0025] Construct an inverse Gaussian pyramid of the selected dynamic image;

[0026] Performing feature fusion on the information scale vector and the inverse Gaussian pyramid to obtain an inverse Gaussian pyramid after feature fusion;

[0027] The reconstructed image corresponding to the selected dynamic image is generated through the inverse Gaussian pyramid after feature fusion.

[0028] In this embodiment, the reconstructed image is block-encoded using a deep learning-based encoder, and the image coding features corresponding to the selected dynamic image are obtained, specifically including:

[0029] Dividing the reconstructed image into blocks to obtain all image sub-blocks;

[0030] Encode each image sub-block according to the encoder based on deep learning to obtain the encoding features corresponding to each image sub-block;

[0031] All coding features are averaged and pooled to obtain the image coding features corresponding to the selected dynamic image.

[0032] In this embodiment, a streaming media transmission protocol is used to transmit real-time data of the image coding features corresponding to the dynamic image at each moment.

[0033] In this embodiment, real-time data transmission of the image coding features corresponding to the dynamic image at each moment specifically includes:

[0034] Packing the image coding features corresponding to the dynamic image at each moment into a data packet;

[0035] The data packet is transmitted to a receiving end based on a streaming media transmission protocol.

[0036] In this embodiment, it also includes: after the receiving end receives the data packet, the data packet is decoded by a deep learning-based decoder.

[0037] The technical solution provided by the embodiments disclosed in this application has the following beneficial effects:

[0038] The image data acquisition unit acquires the dynamic image data stream to be transmitted, and selects the dynamic image in the dynamic image data stream; the scale feature extraction unit performs feature mining on the selected dynamic image to obtain the scale feature vector of the selected dynamic image, and then determines the information singularity corresponding to the scale feature vector, and converts the scale feature vector into the information scale vector of the selected dynamic image based on the information singularity; the image feature encoding unit reconstructs according to the information scale vector to obtain a reconstructed image corresponding to the selected dynamic image, and uses a deep learning-based encoder to block-encode the reconstructed image to obtain the image coding feature corresponding to the selected dynamic image; the coding feature transmission unit continues to select dynamic images in the dynamic image data stream, and then obtains the image coding feature corresponding to the dynamic image at each moment in the dynamic image data stream, and performs real-time data transmission on the image coding feature corresponding to the dynamic image at each moment.

[0039] It can be seen that in the present application, the scale features in the dynamic image data can be effectively extracted and encoded to improve the transmission efficiency of the dynamic image transmission; wherein, by extracting the scale feature vector, the details and structural information at different scales in the dynamic image can be captured, and the calculation of the information singularity helps to identify and quantify the diversity and importance of the features, and by converting the scale feature vector into the information scale vector, more valuable information can be retained in the transmission process to avoid redundant transmission of useless information; then, by reconstructing the image using the information scale vector, it can be ensured that important visual features and structural information are retained during the reconstruction process, and the reconstructed image is processed in blocks, so that the encoder can extract and compress features for each image sub-block, and the encoder based on deep learning has a powerful feature learning ability, which can adaptively extract important features from the data, and can reduce the transmission of redundant information; finally, the streaming media transmission protocol is used to transmit real-time data of the image coding features corresponding to the dynamic image at each moment to ensure the real-time nature of image transmission.

[0040] In summary, the technical solution adopted in the present application can effectively extract and encode scale features in dynamic image data to improve the transmission efficiency of dynamic image transmission. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only the embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0042] Figure 1It is a module structure diagram of a dynamic image transmission system based on deep learning provided by the present application;

[0043] Figure 2 is a schematic diagram of a process for determining information singularity according to the present application;

[0044] Figure 3 It is a schematic diagram of the process of determining image coding features provided by this application. DETAILED DESCRIPTION

[0045] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0046] The embodiment of the present application provides a dynamic image transmission system based on deep learning, the core of which is to collect the dynamic image data stream to be transmitted through the image data collection unit, select the dynamic image in the dynamic image data stream; perform feature mining on the selected dynamic image through the scale feature extraction unit to obtain the scale feature vector of the selected dynamic image, and then determine the information singularity corresponding to the scale feature vector, and convert the scale feature vector into the information scale vector of the selected dynamic image based on the information singularity; the image feature encoding unit reconstructs according to the information scale vector to obtain the reconstructed image corresponding to the selected dynamic image, and uses the encoder based on deep learning to block encode the reconstructed image to obtain the image encoding feature corresponding to the selected dynamic image; the encoding feature transmission unit continues to select the dynamic image in the dynamic image data stream, and then obtains the image encoding feature corresponding to the dynamic image at each moment in the dynamic image data stream, and performs real-time data transmission on the image encoding feature corresponding to the dynamic image at each moment. The present application adopts the above technical solution to effectively extract and encode the scale features in the dynamic image data, so as to improve the transmission efficiency of dynamic image transmission.

[0047] In order to better understand the above technical solution, the above technical solution will be described in detail below in conjunction with the accompanying drawings and specific implementation methods. Figure 1 As shown, the figure is a module structure diagram of a dynamic image transmission system based on deep learning according to the present embodiment of the present application. The processing system includes: an image data acquisition unit 100, a scale feature extraction unit 200, an image feature encoding unit 300, and a coding feature transmission unit 400, which are respectively described as follows:

[0048] The image data acquisition unit 100 acquires a dynamic image data stream to be transmitted and selects a dynamic image from the dynamic image data stream.

[0049] In a specific implementation, the dynamic image data stream to be transmitted can be collected by a mobile camera; wherein the dynamic image data stream represents a series of dynamic images arranged in chronological order.

[0050] It should be noted that in the present application, a dynamic image at one moment in the dynamic image data stream is selected as a processing example. In actual implementation, dynamic images at other moments are processed in the same manner, which will not be repeated here.

[0051] The scale feature extraction unit 200 performs feature mining on the selected dynamic image to obtain a scale feature vector of the selected dynamic image, further determines the information singularity corresponding to the scale feature vector, and converts the scale feature vector into an information scale vector of the selected dynamic image based on the information singularity.

[0052] In this embodiment, feature mining is performed on the selected dynamic image to obtain the scale feature vector of the selected dynamic image in the following manner, namely:

[0053] Gray-scale the selected dynamic image to obtain a gray-scale image corresponding to the selected dynamic image;

[0054] constructing a Gaussian pyramid of a selected dynamic image based on the grayscale image;

[0055] Mining the image of each layer in the Gaussian pyramid to obtain the scale feature of the image of each layer in the Gaussian pyramid;

[0056] The scale feature vector of the selected dynamic image is determined through all the scale features.

[0057] In specific implementation, first, the selected dynamic image is converted into a grayscale image. Grayscale conversion can be achieved by converting the color value of each pixel in the RGB image into a single brightness value, thereby helping to simplify image data, reduce computational complexity, and retain the main brightness information; then, a Gaussian pyramid of the selected dynamic image can be constructed based on the grayscale image. The Gaussian pyramid is a multi-scale image representation method, which is constructed through step-by-step downsampling and Gaussian blur operations. The grayscale image corresponding to the selected dynamic image can be step-by-step downsampled, and Gaussian blur operations are applied to each level of the image to obtain images of different resolutions, and the pyramid composed of all images is used as the Gaussian pyramid of the selected dynamic image; further, for each layer of the Gaussian pyramid of the selected dynamic image, the local binary patterns (Local Binary Patterns, LBP) are used in this application to mine the scale features of each layer of the Gaussian pyramid of the selected video frame; finally, the vector composed of the mined scale features is used as the scale feature vector of the selected dynamic image. In actual implementation, other image feature mining methods can also be used, such as Histogram of Oriented Gradients (Histogram of Oriented Gradients, HOG), etc., which will not be repeated here.

[0058] Preferably, in this embodiment, the information singularity corresponding to the scale feature vector is determined by referring to Figure 2 The figure is a schematic diagram of the process of determining information singularity in some embodiments of the present application. In this embodiment, determining information singularity can be implemented by the following steps:

[0059] First, in step S21, a covariance matrix of the scale feature vector is constructed;

[0060] Then, in step S22, the covariance matrix is ​​subjected to eigenvalue decomposition to obtain a plurality of eigenvalues ​​and eigenvectors corresponding to each eigenvalue;

[0061] Secondly, in step S23, the eigenvectors corresponding to the largest K eigenvalues ​​are selected, and then a feature matrix is ​​constructed through all the selected eigenvectors;

[0062] Finally, in step S24, the information singularity corresponding to the scaled eigenvector is determined by the eigenmatrix.

[0063] In specific implementation, first, a covariance matrix of the scale eigenvector can be constructed, and the covariance matrix of the scale eigenvector contains the correlation information between the scale eigenvectors; then, the covariance matrix is ​​subjected to eigenvalue decomposition to obtain multiple eigenvalues ​​and eigenvectors corresponding to each eigenvalue. By performing eigenvalue decomposition on the covariance matrix of the scale eigenvector, multiple eigenvalues ​​and corresponding eigenvectors of the covariance matrix can be obtained, and the eigenvalue represents the degree of dispersion of the scale eigenvector in the direction of the corresponding eigenvector; secondly, the eigenvectors corresponding to the largest k eigenvalues ​​are selected, and the characteristic matrix corresponding to the scale eigenvector can be constructed through these eigenvectors; finally, the information singularity corresponding to the scale eigenvector can be determined through the characteristic matrix, that is, the matrix value of the characteristic matrix can be used as the information singularity. It should be noted that the value of k can be determined based on historical experience. In this application, the information singularity represents the degree of difference in the information distribution of the scale eigenvector.

[0064] In this embodiment, the scale feature vector is converted into the information scale vector of the selected dynamic image based on the information singularity in the following manner, namely:

[0065] constructing an orthogonal matrix of the scale eigenvectors;

[0066] The scale feature vector is transformed using the information singularity and the orthogonal matrix to obtain the information scale vector of the selected dynamic image.

[0067] In the specific implementation, first, an orthogonal matrix of the scale feature vector is constructed. The principal component analysis (PCA) dimensionality reduction technique can be used on the scale feature vector to obtain a reduced-dimensional feature matrix, which is used as the orthogonal matrix corresponding to the scale feature vector. Then, the scale feature vector can be transformed using the information singularity and the orthogonal matrix to obtain the information scale vector of the selected dynamic image. In the actual implementation, the information scale vector can be determined according to the following formula:

[0068]

[0069] in, represents the information scale vector, E represents the orthogonal matrix corresponding to the scale feature vector, E T represents the transposed matrix of the orthogonal matrix, H represents the information singularity, and F represents the scale feature vector. It should be noted that in the present application, the information scale vector is a scale feature vector obtained by removing the multicollinearity between features through feature scaling.

[0070] It should be noted that by extracting the scale feature vector, the details and structural information at different scales in the dynamic image can be captured, and the calculation of information singularity helps to identify and quantify the diversity and importance of features. By converting the scale feature vector into the information scale vector, more valuable information can be retained in the transmission process, avoiding redundant transmission of useless information.

[0071] The image feature encoding unit 300 reconstructs according to the information scale vector to obtain a reconstructed image corresponding to the selected dynamic image, and uses a deep learning-based encoder to block-encode the reconstructed image to obtain image coding features corresponding to the selected dynamic image.

[0072] In this embodiment, the following method can be used to reconstruct the selected dynamic image according to the information scale vector, namely:

[0073] Construct an inverse Gaussian pyramid of the selected dynamic image;

[0074] Performing feature fusion on the information scale vector and the inverse Gaussian pyramid to obtain an inverse Gaussian pyramid after feature fusion;

[0075] The reconstructed image corresponding to the selected dynamic image is generated through the inverse Gaussian pyramid after feature fusion.

[0076] In specific implementation, first, the inverse operation can be used to construct an inverse Gaussian pyramid of the selected dynamic image; then, the information scale vector and the inverse Gaussian pyramid can be feature fused, that is, the information scale vector is added to the image of each layer in the inverse Gaussian pyramid, so as to obtain the inverse Gaussian pyramid after feature fusion; finally, the reconstructed image corresponding to the selected dynamic image can be generated by the inverse Gaussian pyramid after feature fusion, that is, the image corresponding to the top layer in the inverse Gaussian pyramid after feature fusion is selected as the reconstructed image corresponding to the selected dynamic image.

[0077] Preferably, in this embodiment, a deep learning-based encoder is used to perform block encoding on the reconstructed image, thereby obtaining image encoding features corresponding to the selected dynamic image, referring to Figure 3 The figure is a schematic diagram of a process for determining image coding features in some embodiments of the present application. In this embodiment, determining image coding features can be implemented by the following steps:

[0078] First, in step S31, the reconstructed image is divided into blocks to obtain all image sub-blocks;

[0079] Then, in step S32, each image sub-block is encoded according to the encoder based on deep learning to obtain the encoding features corresponding to each image sub-block;

[0080] Finally, in step S33, all coding features are average pooled to obtain image coding features corresponding to the selected dynamic image.

[0081] In the specific implementation, first, the reconstructed image can be divided into blocks to obtain all image sub-blocks, and a suitable block size (such as 8x8, 16x16 or 32x32 sub-blocks) can be selected to traverse the reconstructed image and divide it into multiple non-overlapping image sub-blocks; then, each image sub-block can be encoded separately according to the deep learning-based encoder, and a suitable deep learning encoder architecture (such as convolutional neural network, autoencoder, etc.) can be selected. The encoder needs to be trained to effectively extract image features, and each image sub-block is input into the encoder. The encoder extracts and compresses features and outputs corresponding coding features; finally, all coding features can be averaged and pooled to obtain image coding features corresponding to the selected dynamic image, that is, the values ​​of all coding features are averaged to obtain a unified feature vector, that is, the image coding features corresponding to the selected dynamic image, and the image coding features represent the encoded image features corresponding to the selected dynamic image.

[0082] It should be noted that by using the information scale vector to reconstruct the image, it can be ensured that important visual features and structural information are retained during the reconstruction process, and the reconstructed image is processed in blocks, so that the encoder can extract and compress features for each image sub-block. The encoder based on deep learning has powerful feature learning capabilities and can adaptively extract important features from the data, which can reduce the transmission of redundant information.

[0083] The coding feature transmission unit 400 continues to select dynamic images in the dynamic image data stream, and then obtains the image coding features corresponding to the dynamic images at each moment in the dynamic image data stream, and performs real-time data transmission on the image coding features corresponding to the dynamic images at each moment.

[0084] In specific implementation, dynamic images can continue to be selected in the dynamic image data stream, and then the continuously selected dynamic images can be processed by the method of the present application to obtain the image coding features corresponding to the dynamic images at each moment in the dynamic image data stream.

[0085] In this embodiment, a streaming media transmission protocol is used to transmit real-time data of the image coding features corresponding to the dynamic image at each moment.

[0086] In this embodiment, real-time data transmission of the image coding features corresponding to the dynamic image at each moment specifically includes:

[0087] Packing the image coding features corresponding to the dynamic image at each moment into a data packet;

[0088] The data packet is transmitted to a receiving end based on a streaming media transmission protocol.

[0089] In specific implementation, the image coding features corresponding to the dynamic image at each moment can be packaged into a data packet, that is, the image coding features are converted into a format suitable for transmission, such as JSON, protobuf or a custom binary format. This step ensures that the image coding features can be correctly parsed and processed during the transmission process, so that the formatted image coding features are encapsulated into a data packet to ensure that the data structure is complete and easy to transmit; then, a suitable streaming media transmission protocol can be selected according to actual needs. The streaming media transmission protocol selected in this application is RTSP (Real-Time Streaming Protocol). Before transmission, a handshake process is used to confirm the connection between the client and the receiver to ensure that the data can be transmitted smoothly, and the packaged data packet is sent to the receiver through the selected protocol.

[0090] In this embodiment, it also includes: after the receiving end receives the data packet, the data packet is decoded by a deep learning-based decoder; in the specific implementation, the corresponding deep learning decoder is selected, and the received data packet is input into the decoder for decoding. The corresponding coding features can be obtained through the decoding process, so that the coding features can be converted into original image data.

[0091] It can be seen that in the present application, the scale features in the dynamic image data can be effectively extracted and encoded to improve the transmission efficiency of the dynamic image transmission; wherein, by extracting the scale feature vector, the details and structural information at different scales in the dynamic image can be captured, and the calculation of the information singularity helps to identify and quantify the diversity and importance of the features, and by converting the scale feature vector into the information scale vector, more valuable information can be retained in the transmission process to avoid redundant transmission of useless information; then, by reconstructing the image using the information scale vector, it can be ensured that important visual features and structural information are retained during the reconstruction process, and the reconstructed image is processed in blocks, so that the encoder can extract and compress features for each image sub-block, and the encoder based on deep learning has a powerful feature learning ability, which can adaptively extract important features from the data, and can reduce the transmission of redundant information; finally, the streaming media transmission protocol is used to transmit real-time data of the image coding features corresponding to the dynamic image at each moment to ensure the real-time nature of image transmission.

[0092] In summary, the technical solution adopted in the present application can effectively extract and encode scale features in dynamic image data to improve the transmission efficiency of dynamic image transmission.

[0093] The present application is described with reference to flowcharts and / or block diagrams of methods, devices (systems) and computer program products according to embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0094] A person skilled in the art can understand that all or part of the steps in the various methods of the above embodiments can be completed by instructing related hardware through a program, and the program can be stored in a computer-readable storage medium, and the storage medium includes a read-only memory (ROM), a random access memory (RAM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), a one-time programmable read-only memory (OTPROM), an electronically erasable rewritable read-only memory (EEPROM), a compact disc (CD-ROM) or other optical disc storage, magnetic disk storage, magnetic tape storage, or any other computer-readable medium that can be used to carry or store data.

[0095] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

Claims

1. A dynamic image transmission system based on deep learning, characterized in that: The transmission system comprises: An image data acquisition unit, which acquires a dynamic image data stream to be transmitted and selects a dynamic image from the dynamic image data stream; A scale feature extraction unit performs feature mining on the selected dynamic image to obtain a scale feature vector of the selected dynamic image, further determines the information singularity corresponding to the scale feature vector, and converts the scale feature vector into an information scale vector of the selected dynamic image based on the information singularity; An image feature encoding unit, reconstructing according to the information scale vector to obtain a reconstructed image corresponding to the selected dynamic image, and using a deep learning-based encoder to block-encode the reconstructed image to obtain an image coding feature corresponding to the selected dynamic image; The coding feature transmission unit continues to select dynamic images in the dynamic image data stream, and then obtains the image coding features corresponding to the dynamic images at each moment in the dynamic image data stream, and performs real-time data transmission on the image coding features corresponding to the dynamic images at each moment.

2. A dynamic image transmission system based on deep learning as claimed in claim 1, characterized in that: The dynamic image data stream to be transmitted is collected by a mobile camera.

3. A dynamic image transmission system based on deep learning as claimed in claim 1, characterized in that: Perform feature mining on the selected dynamic image to obtain the scale feature vector of the selected dynamic image, which specifically includes: Gray-scale the selected dynamic image to obtain a gray-scale image corresponding to the selected dynamic image; constructing a Gaussian pyramid of a selected dynamic image based on the grayscale image; Mining the image of each layer in the Gaussian pyramid to obtain the scale feature of the image of each layer in the Gaussian pyramid; The scale feature vector of the selected dynamic image is determined through all the scale features.

4. A dynamic image transmission system based on deep learning as claimed in claim 1, characterized in that: Determining the information singularity corresponding to the scale feature vector specifically includes: Constructing a covariance matrix of the scaled eigenvector; Performing eigenvalue decomposition on the covariance matrix to obtain multiple eigenvalues ​​and eigenvectors corresponding to each eigenvalue; Select the eigenvectors corresponding to the largest K eigenvalues, and then construct a feature matrix through all the selected eigenvectors; The information singularity corresponding to the scale eigenvector is determined by the eigenmatrix.

5. The deep learning-based dynamic image transmission system according to claim 1, characterized in that: Converting the scale feature vector into the information scale vector of the selected dynamic image based on the information singularity specifically includes: constructing an orthogonal matrix of the scale eigenvectors; The scale feature vector is transformed using the information singularity and the orthogonal matrix to obtain the information scale vector of the selected dynamic image.

6. A dynamic image transmission system based on deep learning as claimed in claim 1, characterized in that: Reconstructing according to the information scale vector to obtain a reconstructed image corresponding to the selected dynamic image specifically includes: Construct an inverse Gaussian pyramid of the selected dynamic image; Performing feature fusion on the information scale vector and the inverse Gaussian pyramid to obtain an inverse Gaussian pyramid after feature fusion; The reconstructed image corresponding to the selected dynamic image is generated through the inverse Gaussian pyramid after feature fusion.

7. A dynamic image transmission system based on deep learning as claimed in claim 1, characterized in that: The reconstructed image is block-encoded using a deep learning-based encoder, and the image coding features corresponding to the selected dynamic image are obtained, specifically including: Dividing the reconstructed image into blocks to obtain all image sub-blocks; Encode each image sub-block according to the encoder based on deep learning to obtain the encoding features corresponding to each image sub-block; All coding features are averaged and pooled to obtain the image coding features corresponding to the selected dynamic image.

8. The deep learning-based dynamic image transmission system according to claim 1, characterized in that: The streaming media transmission protocol is used to transmit real-time data of the image coding features corresponding to the dynamic image at each moment.

9. The deep learning-based dynamic image transmission system according to claim 1, characterized in that: The real-time data transmission of the image coding features corresponding to the dynamic image at each moment specifically includes: Packing the image coding features corresponding to the dynamic image at each moment into a data packet; The data packet is transmitted to a receiving end based on a streaming media transmission protocol.

10. A dynamic image transmission system based on deep learning as claimed in claim 1, characterized in that: Also includes: After receiving the data packet, the receiving end decodes the data packet through a deep learning-based decoder.

Images