Unidirectional optical shutter video stream transmission method based on weak supervision semantic segmentation

Abstract

The invention discloses a one-way optical shutter video stream transmission method based on weak supervision semantic segmentation, and belongs to the technical field of one-way optical shutter information transmission. Comprising the following steps: carrying out key region and background region classification and marking on all video data received by an optical shutter outer end through a weak supervision semantic segmentation model; performing priority division on a key region and a background region in the video frame; dividing the data packets according to time windows, and packaging the data packets; performing dynamic data compression and segmented coding on the packaged data packet; reed-Solomon redundant coding is carried out on all the data packets of the key area after compression coding is carried out on the data packets of the key area; dynamically adjusting the transmission sequence of the data packets to perform link scheduling; and a receiving end carries out de-encapsulation and data restoration, and reconstructs the restored video frame. According to the method, redundancy and compression of data transmission contents are accurately controlled, so that the transmission efficiency and reliability are remarkably improved, the bandwidth utilization is optimized, and the transmission reliability and stability are enhanced.

Description

Technical Field

[0001] The invention relates to a transmission method of a unidirectional light gate video stream, in particular to a unidirectional light gate video stream transmission method based on weakly supervised semantic segmentation, and belongs to the technical field of unidirectional light gate information transmission. Background Art

[0002] With the rapid development of computer vision technology, weakly supervised semantic segmentation has gradually become an important research direction in image and video processing, especially in the field of video streams. Traditional semantic segmentation methods rely on a large amount of labeled data for training, but the acquisition of labeled data usually requires a lot of labor costs and time. Weakly supervised semantic segmentation technology can significantly reduce the need for data labeling while ensuring accuracy by using a small amount of labeled data and a large amount of unlabeled data for training. This technology can automatically identify and segment key target areas in the video (such as people, vehicles, objects, etc.), and can perform target recognition and region segmentation based on contextual information in the image.

[0003] At present, in video streaming transmission, the combination of weakly supervised semantic segmentation technology has the following advantages: (1) Improve data processing efficiency: Weakly supervised semantic segmentation can effectively extract key areas in the video, and by optimizing bandwidth allocation, reduce the transmission of unimportant background data, thereby improving transmission efficiency. (2) Reduce bandwidth consumption: By intelligently identifying key areas in the video frame, only these important areas are transmitted with high quality, while the background area is processed using a high compression algorithm, which significantly reduces bandwidth usage.

[0004] However, although the one-way optical gate (FGAP) technology has shown unique advantages in network security and data protection, its application in video streaming still faces many challenges:

[0005] 1) Low transmission efficiency: Due to the lack of feedback mechanism, redundant transmission and repeated transmission seriously occupy bandwidth resources;

[0006] 2) Bandwidth waste: Redundant data occupies a large amount of bandwidth, resulting in low bandwidth utilization, especially when bandwidth resources are tight;

[0007] (3) Poor transmission reliability: Data loss or delay will affect the continuity of the video stream and reduce the user experience.

[0008] To address the above problems, it is urgent to design a unidirectional light-gate video stream transmission method based on weakly supervised semantic segmentation, which can improve both transmission efficiency and reliability. Summary of the invention

[0009] The purpose of the present invention is to solve the problems of low transmission efficiency, low bandwidth utilization and poor transmission reliability in existing transmission methods, and to provide a unidirectional optical shutter video stream transmission method based on weakly supervised semantic segmentation, which combines weakly supervised semantic segmentation and is based on intelligent flow control and adaptive optimization, and significantly improves transmission efficiency and reliability and optimizes bandwidth utilization by precisely controlling the redundancy and compression of data transmission content.

[0010] To achieve the above object, the present invention adopts the following technical solution: a one-way shutter video stream transmission method based on weakly supervised semantic segmentation, comprising the following steps:

[0011] S1. Video stream data reception and preprocessing:

[0012] The outer end of the optical gate receives video stream data sent from the external network. First, it uses the pre-trained weakly supervised semantic segmentation model to process part of the received video stream, automatically identifies the key area and background area in the video for each frame of video data, and marks and classifies them.

[0013] Then, the weakly supervised semantic segmentation model is used to extract video frames from all video data received at the outer end of the optical gate, convert them into image data and input them into the trained model. Finally, all received data are classified and marked for key areas and background areas.

[0014] S2. Data prioritization and redundant transmission:

[0015] Prioritize the key area and background area in the classified video frame, assign high priority to the key area data stream, generate key area redundant data packets by adding redundant data and performing protocol stripping, assign low priority to the background area data stream, and perform protocol stripping to generate background area data packets, and record different data packet numbers and corresponding task IDs respectively;

[0016] S3, time window cache:

[0017] The constructed key area redundant data packets and background area data packets are divided into time windows respectively, the data in each window is processed and cached independently, the video frames in each time window are classified and processed according to the key area and the background area, and the processed key area redundant data packets and background area data packets are respectively encapsulated into data packets using their own packetization protocols;

[0018] S4. Data compression and encoding:

[0019] According to the real-time bandwidth situation, the H.265 compression algorithm is used to dynamically compress the encapsulated key area redundant data packets and background area data packets, and the compression rate of the video frame data is dynamically adjusted. A low QP value of 0 to 20 is used for the key area, and a high QP value of 31 to 51 is used for the background area, while maintaining basic image quality;

[0020] Each frame of video data is segmented and encoded, and each segment of data is transmitted independently. The H.265 high-quality encoding strategy is used for video frames in key areas. When bandwidth is sufficient, clear target areas are transmitted first. A high QP value encoding method is used for background areas to ensure maximum bandwidth utilization.

[0021] S5. Add asynchronous error correction coding:

[0022] Perform Reed-Solomon redundancy coding on all key area data packets after compression coding, and transmit the redundant data and error correction code together with the original video data to ensure that key data can be restored through redundant data even if the packet is lost;

[0023] S6. Link Scheduling:

[0024] The redundant data packets of the key area and the data packets of the background area are sequentially added to the sending queue channel to wait for sending. According to the transmission status and bandwidth occupancy, the transmission order of the data packets is dynamically adjusted, the key area data is transmitted first, and the high priority background data is transmitted secondly, and the data packets are sent to the inner end of the one-way optical gate through one-way transmission;

[0025] S7, receiving end decapsulation and data restoration:

[0026] The redundant data packets of the key area and the data packets of the background area arriving at the inner end of the one-way optical gate are stored in the receiving queue channel. The receiving end decapsulates each received data packet according to the protocol corresponding to the packet according to the task ID and identification information of each data packet, and restores the data of the key area and the data of the background area respectively;

[0027] The restored video frames are reconstructed, and the cached background area and the key area data transmitted in real time are reconstructed according to the received data. The missing parts of the key area are restored through redundant data to ensure smooth playback of the video.

[0028] In step S1, the weakly supervised semantic segmentation model adopts a VisionTransformer (ViT) model based on a convolutional neural network.

[0029] In step S1, the key areas are objects that the user is concerned about, including faces, vehicles and key objects; the background areas are unimportant areas or areas that do not need to be displayed in detail.

[0030] In step S3, the time window is divided into every 5 seconds or 10 seconds as a time window.

[0031] In the step S4, the H.265 high-quality coding strategy is divided into a video coding layer and a network extraction layer in the coding structure. The coding process of the video coding layer is as follows: 1) divide the image of each frame into blocks and add the block information to the code stream; 2) perform predictive coding on the unit block, generate residuals by intra-frame prediction, and perform motion estimation and motion compensation by inter-frame prediction; 3) transform the residual, quantize and scan the transform coefficients; 4) perform entropy coding on the quantized transform coefficients, motion information, and prediction information to form a compressed video code stream output; the network extraction layer mainly defines the data encapsulation format, encapsulates the video data generated by the video coding layer into data packets of network extraction layer units, adapts to different network environments and transmits them.

[0032] In step S5, the encoding process of Reed-Solomon redundant coding is as follows: 1) Input data vector: The input data is regarded as a vector D=(D1, D2, ..., D n ), there are n data symbols;

[0033] 2) Coding Matrix: Define the coding matrix B, which is an n×n matrix used to combine the input data with the redundant symbols. The encoded data vector is (D, C) = (D1, D2, …, D n ,C1,C2,…,C m ), where C1, C2, …, C m is a redundant symbol;

[0034] In matrix operations, the encoding process is expressed as: Y=B×D, where Y is the encoded data, Y includes original data and redundant symbols, D is the original data vector, and B is the encoding matrix.

[0035] In step S5, the error correction and data recovery process of Reed-Solomon redundant coding is as follows: 1) Identification of lost data blocks: Assuming that the data vector received by the receiving end is R, some symbols may be lost or erroneous, by comparing the received data vector with the coding matrix, a survivors vector is constructed to represent the correct symbols received;

[0036] 2) Construct submatrix and recover data: The error correction and data recovery process is as follows: the result after encoding can tolerate at most m data block errors, and the encoding matrix corresponding to the remaining correct data blocks is taken out to form an n*n submatrix, denoted as B'. The survivors vector represents the information received by the receiver or indicates that D1, D4, C2 packets are lost, and its corresponding matrix is ​​D. According to the relevant theorem of the matrix, it can be known that the inverse matrix B' of B' is obtained-1 , we can get D = B' -1 *survivors.

[0037] The beneficial effects of the present invention are:

[0038] 1) The present invention combines weakly supervised semantic segmentation with intelligent flow control and adaptive optimization, and significantly improves transmission efficiency and reliability and optimizes bandwidth utilization by precisely controlling the redundancy and compression of data transmission content.

[0039] 2) By combining the simulated feedback mechanism and weakly supervised semantic segmentation, the present invention can analyze the key target areas in the video stream in real time during the transmission process, and only perform redundant transmission and high-quality processing on these key areas, while using compression and low-redundancy strategies for background areas, reducing redundant transmission and unnecessary bandwidth waste, thereby significantly improving transmission efficiency, optimizing bandwidth usage, and improving bandwidth utilization.

[0040] 3) The present invention combines intelligent packet loss prediction, asynchronous error correction and adaptive compression technology to ensure the timeliness of video stream transmission while using weakly supervised semantic segmentation technology to ensure the integrity of key data areas and enhance the reliability and stability of transmission.

[0041] 4) The present invention can flexibly adapt to changes in the network environment, automatically adjust bandwidth allocation and transmission strategies, and adapt to various video formats and bandwidth fluctuations through dynamic flow control and adaptive optimization algorithms.

[0042] 5) The present invention reduces the complex protocols and redundant mechanisms in traditional systems by introducing intelligent prediction and error correction mechanisms, reduces the complexity of system design and operation and maintenance, and improves the stability and maintainability of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 is an overall flow chart of the transmission method of the present invention;

[0044] Figure 2 It is a framework flow chart of the weakly supervised semantic segmentation model used in the transmission method of the present invention;

[0045] Figure 3 This is a framework flow chart of H.265 high-quality encoding used in the transmission method of the present invention;

[0046] Figure 4 The schematic diagram is a schematic diagram of the Reed-Solomon encoding used in the transmission method of the present invention. DETAILED DESCRIPTION

[0047] The present invention is further explained below in conjunction with the accompanying drawings and specific embodiments.

[0048] Example: Figure 1-4 As shown, the present invention provides a one-way shutter video stream transmission method based on weakly supervised semantic segmentation, comprising the following steps:

[0049] S1. Video stream data reception and preprocessing:

[0050] The outer end of the optical gate receives video stream data sent from the external network, and first uses the pre-trained weakly supervised semantic segmentation model to process part of the received video stream. The weakly supervised semantic segmentation model adopts the Vision Transformer (ViT) model based on convolutional neural networks. For each frame of video data, it automatically identifies the key areas and background areas in the video and marks and classifies them; the key areas are the objects that the user pays attention to, including faces, vehicles and key objects; the background areas are unimportant areas or areas that do not require detailed display.

[0051] Then, a weakly supervised semantic segmentation model is used to extract video frames from all video data received at the outer end of the optical gate, convert them into image data and input them into the trained model. Finally, all received data are classified and marked for key areas and background areas.

[0052] The basic framework of the weakly supervised semantic segmentation model is as follows Figure 2 As shown in , weakly supervised semantic segmentation methods usually adopt a two-stage process: the first stage generates preliminary seed regions through class activation maps (CAM) to locate the approximate position of the object; the second stage optimizes these seed regions and generates high-quality pixel-level pseudo labels through refinement and optimization techniques (such as CRF). These pseudo labels can be used to train the segmentation model to achieve more accurate semantic segmentation without complete pixel-level annotation. This method effectively reduces the cost of manual annotation while maintaining high segmentation accuracy.

[0053] S2. Data prioritization and redundant transmission:

[0054] Priority is given to the key areas and background areas in the classified video frames, where the key area data stream is given high priority. Redundant data packets in the key areas are generated by adding redundant data and performing protocol stripping to ensure sufficient bandwidth and quality during transmission.

[0055] The background area data stream is assigned a low priority, and then the protocol is stripped to generate the background area data packet, and the sequence numbers of different data packets and the corresponding task IDs are recorded respectively.

[0056] Redundant data transmission: For key areas, redundant data is added to improve transmission reliability and reduce the risk of packet loss in key areas; the amount of redundant transmission is dynamically adjusted by predicting packet loss risks and analyzing network bandwidth conditions.

[0057] S3, time window cache:

[0058] The constructed key area redundant data packets and background area data packets are divided into time windows respectively, for example, 5 seconds or 10 seconds are divided into a time window, the data in each window is processed and cached independently, the video frames in each time window are classified and processed according to the key area and the background area, and the processed key area redundant data packets and background area data packets are encapsulated into data packets using their own packetization protocols.

[0059] Transmission strategy adjustment: If bandwidth resources are sufficient, data in key areas will be transmitted first, and background data can be delayed or redundant transmission can be reduced; when bandwidth is tight, the system will automatically reduce the transmission frequency of background areas and appropriately increase the redundancy of key areas to ensure the continuity of key data.

[0060] S4. Data compression and encoding:

[0061] According to the real-time bandwidth situation, the H.265 compression algorithm is used to dynamically compress the encapsulated redundant data packets in the key area and the background area data packets, and the compression rate of the video frame data is dynamically adjusted. A low QP value of 0 to 20 is used for the key area to ensure high-quality transmission; a high QP value of 31 to 51 is used for the background area while maintaining basic image quality.

[0062] Each frame of video data is segmented and encoded, and the data of each segment is transmitted independently. Even if some segments are lost, the data of other segments can still be completely decoded to ensure the continuity of playback.

[0063] Coding optimization: Figure 3 As shown in the figure, the H.265 high-quality encoding strategy is used for the video frames in the key area, and the clear target area is transmitted first when the bandwidth is sufficient; the encoding method with a high QP value of 31 to 51 is used for the background area to ensure maximum bandwidth utilization.

[0064] The H.265 high-quality coding strategy is divided into a video coding layer and a network extraction layer in terms of coding structure. The coding process of the video coding layer is as follows: 1) Divide the image of each frame into blocks and add the block information to the bitstream; 2) Perform predictive coding on the unit blocks, generate residuals through intra-frame prediction, and perform motion estimation and motion compensation through inter-frame prediction; 3) Transform the residuals, quantize and scan the transform coefficients; 4) Perform entropy coding on the quantized transform coefficients, motion information, and prediction information to form a compressed video bitstream output; The network extraction layer mainly defines the data encapsulation format, encapsulates the video data generated by the video coding layer into data packets of network extraction layer units, adapts to different network environments and transmits them.

[0065] The overall encoding process of H.265 includes several key steps. First, the input video frame is divided into coding units (CU) of different sizes. Each CU can be further subdivided into prediction units (PU) and transform units (TU) according to the content. In the inter-frame prediction process, multiple reference frames and more sophisticated motion estimation techniques are used to generate motion vectors through precise motion compensation. Next, intra-frame prediction is performed, and H.265 uses 35 prediction modes to optimize the encoding of static areas. Subsequently, transformation and quantization are performed to convert image data from the time domain to the frequency domain to reduce redundant information and compress data. The transformed data is adaptively quantized and entropy encoded using CABAC to further improve compression efficiency. Finally, a compressed bitstream is generated, which contains all the encoding information of the video frame. The entire process significantly improves video compression rate and encoding quality through flexible block division, sophisticated motion compensation and efficient encoding strategies.

[0066] S5. Add asynchronous error correction coding:

[0067] Reed-Solomon redundancy coding is performed on all key area data packets after compression coding, and the redundant data and error correction code are transmitted together with the original video data to ensure that key data can be restored through redundant data even if packets are lost.

[0068] Reed-Solomon redundancy coding uses the characteristics of the Van der Leyen matrix or the Cauchy matrix to realize the function of error correction code. Its principle is as follows Figure 4 shown.

[0069] The encoding process is as follows Figure 4 (a) shown.

[0070] 1) Input data vector: Consider the input data as a vector D = (D1, D2, ..., D n ), there are n data symbols.

[0071] 2) Coding Matrix: Define the coding matrix B, which is an n×n matrix used to combine the input data with the redundant symbols. The encoded data vector is (D, C) = (D1, D2, …, D n ,C1,C2,…,C m ), where C1, C2, …, C m is a redundant symbol.

[0072] In matrix operations, the encoding process is expressed as: Y=B×D, where Y is the encoded data (including original data and redundant symbols), D is the original data vector, and B is the encoding matrix.

[0073] Error correction and data recovery process Figure 4 As shown in (b) and (c).

[0074] 1) Identification of lost data blocks: Assume that the data vector received by the receiver is R, in which some symbols may be lost or erroneous. In order to recover the data, the receiver first needs to know which symbols are lost. By comparing the received data vector with the encoding matrix, a survivors vector can be constructed to represent the correct symbols received.

[0075] 2) Construct submatrix and recover data: The result after encoding can tolerate at most m data block errors. The encoding matrix corresponding to the remaining correct data blocks is taken out to form an n*n submatrix, denoted as B'. The survivors vector represents the information received by the receiver or indicates that D1, D4, C2 packets are lost. Its corresponding matrix is ​​D. According to the relevant theorem of the matrix, we can find the inverse matrix B' of B' -1 , we can get D = B' -1 *survivors.

[0076] S6. Link Scheduling:

[0077] The redundant data packets in the key area and the data packets in the background area are added to the sending queue channel in turn and wait for transmission. The transmission order of the data packets is dynamically adjusted according to the transmission status and bandwidth occupancy, with priority given to the transmission of the key area data and second priority given to the transmission of the high-priority background data. When the bandwidth resources change, the system will dynamically adjust the transmission order of the data packets according to the complexity and priority of the video content; and send them to the inner end of the unidirectional optical gate through unidirectional transmission.

[0078] S7, receiving end decapsulation and data restoration:

[0079] The redundant data packets of the key area and the background area that arrive at the inner end of the one-way optical gate are stored in the receiving queue channel. The receiving end decapsulates each received data packet according to the protocol corresponding to its packet based on the task ID and identification information of each data packet, and restores the key area data and the background area data respectively; for each video frame, the system decodes the data of the key area and the background area to ensure the integrity of the picture after decoding.

[0080] The restored video frames are reconstructed, and the cached background area and the key area data transmitted in real time are reconstructed according to the received data to ensure smooth playback of the video; the missing parts of the key area are restored through redundant data to avoid interruptions or freezes during playback.

[0081] Working principle:

[0082] The present invention utilizes the physical isolation characteristics of the one-way optical gate and combines it with weakly supervised semantic segmentation technology to achieve efficient, safe and stable transmission of video streams through the following aspects.

[0083] 1) Data preparation and priority division: After receiving the video stream data, each frame of the image is analyzed through weakly supervised semantic segmentation to identify and mark key target areas (such as people, vehicles, objects, etc.); these important areas are then marked as high-priority data for key processing. For unimportant background data, its priority is lowered, and compression and redundancy strategies are used to reduce bandwidth usage.

[0084] 2) Compression and segmented encoding: The compression rate is adjusted dynamically according to the real-time bandwidth conditions and data priority. A low compression rate is used for critical data to ensure quality, while a high compression rate is used for background data to reduce bandwidth consumption. At the same time, the compressed video data is segmented to ensure the independence of each segment of data and avoid the loss of a single segment affecting the overall video playback.

[0085] 3) Time window division and caching: Video data is divided into time windows, and the data in each time window is cached. Edge nodes are used to cache data, non-urgent data is loaded in advance, and the data required for real-time playback is kept in transmission; combined with weakly supervised semantic segmentation, the integrity of key data areas is ensured.

[0086] 4) Error correction and redundancy mechanism: In data transmission, asynchronous error correction codes are used to perform fault-tolerant encoding on key data packets and increase redundancy. Combined with weakly supervised semantic segmentation technology, additional redundancy is added to key data areas to ensure that even if packet loss occurs, key data can be automatically repaired to avoid retransmission.

[0087] 5) Link scheduling and dynamic feedback: Redundancy and data priority are adjusted in real time through proxy servers and simulation feedback mechanisms. Bandwidth usage is dynamically adjusted based on real-time transmission results to ensure that critical data is transmitted first when bandwidth resources are limited.

[0088] The present invention combines weakly supervised semantic segmentation with intelligent flow control and adaptive optimization, and significantly improves transmission efficiency and reliability and optimizes bandwidth utilization by precisely controlling the redundancy and compression of data transmission content. At the same time, it combines intelligent packet loss prediction, asynchronous error correction and adaptive compression technology, and uses weakly supervised semantic segmentation technology to ensure the integrity of key data areas while ensuring the timeliness of video stream transmission, thereby enhancing the reliability and stability of transmission.

[0089] The above description is only used to illustrate the technical solution of the present invention rather than to limit it. Other modifications or equivalent substitutions made to the technical solution of the present invention by ordinary technicians in this field should be included in the scope of the claims of the present invention as long as they do not depart from the spirit and scope of the technical solution of the present invention.

Claims

1. A one-way shutter video stream transmission method based on weakly supervised semantic segmentation, characterized in that: The following steps are involved: S1. Video stream data reception and preprocessing: The outer end of the optical gate receives video stream data sent from the external network. First, it uses the pre-trained weakly supervised semantic segmentation model to process part of the received video stream, automatically identifies the key area and background area in the video for each frame of video data, and marks and classifies them. Then, the weakly supervised semantic segmentation model is used to extract video frames from all video data received at the outer end of the optical gate, convert them into image data and input them into the trained model. Finally, all received data are classified and marked for key areas and background areas. S2. Data prioritization and redundant transmission: Prioritize the key area and background area in the classified video frame, assign high priority to the key area data stream, generate key area redundant data packets by adding redundant data and performing protocol stripping, assign low priority to the background area data stream, and perform protocol stripping to generate background area data packets, and record different data packet numbers and corresponding task IDs respectively; S3, time window cache: The constructed key area redundant data packets and background area data packets are divided into time windows respectively, the data in each window is processed and cached independently, the video frames in each time window are classified and processed according to the key area and the background area, and the processed key area redundant data packets and background area data packets are respectively encapsulated into data packets using their own packetization protocols; S4. Data compression and encoding: According to the real-time bandwidth situation, the H.265 compression algorithm is used to dynamically compress the encapsulated key area redundant data packets and background area data packets, and the compression rate of the video frame data is dynamically adjusted. A low QP value of 0 to 20 is used for the key area, and a high QP value of 31 to 51 is used for the background area, while maintaining basic image quality; Each frame of video data is segmented and encoded, and each segment of data is transmitted independently. The H.265 high-quality encoding strategy is used for video frames in key areas. When bandwidth is sufficient, clear target areas are transmitted first. A high QP value encoding method is used for background areas to ensure maximum bandwidth utilization. S5. Add asynchronous error correction coding: Perform Reed-Solomon redundancy coding on all key area data packets after compression coding, and transmit the redundant data and error correction code together with the original video data to ensure that key data can be restored through redundant data even if the packet is lost; S6. Link Scheduling: The redundant data packets of the key area and the data packets of the background area are sequentially added to the sending queue channel to wait for sending. According to the transmission status and bandwidth occupancy, the transmission order of the data packets is dynamically adjusted, the key area data is transmitted first, and the high priority background data is transmitted secondly, and the data packets are sent to the inner end of the one-way optical gate through one-way transmission; S7, receiving end decapsulation and data restoration: The redundant data packets of the key area and the data packets of the background area arriving at the inner end of the one-way optical gate are stored in the receiving queue channel. The receiving end decapsulates each received data packet according to the protocol corresponding to the packet according to the task ID and identification information of each data packet, and restores the data of the key area and the data of the background area respectively; The restored video frames are reconstructed, and the cached background area and the key area data transmitted in real time are reconstructed according to the received data. The missing parts of the key area are restored through redundant data to ensure smooth playback of the video.

2. According to claim 1, a one-way shutter video stream transmission method based on weakly supervised semantic segmentation is characterized in that: In step S1, the weakly supervised semantic segmentation model adopts a VisionTransformer (ViT) model based on a convolutional neural network.

3. The one-way shutter video stream transmission method based on weakly supervised semantic segmentation according to claim 1 is characterized in that: In step S1, the key areas are objects that the user is concerned about, including faces, vehicles and key objects; the background areas are unimportant areas or areas that do not need to be displayed in detail.

4. The one-way shutter video stream transmission method based on weakly supervised semantic segmentation according to claim 1 is characterized in that: In step S3, the time window is divided into every 5 seconds or 10 seconds as a time window.

5. The one-way shutter video stream transmission method based on weakly supervised semantic segmentation according to claim 1 is characterized in that: In step S4, the H.265 high-quality coding strategy is divided into a video coding layer and a network extraction layer in the coding structure, and the coding process of the video coding layer is as follows: 1) the image of each frame is divided into blocks, and the block information is added to the code stream; 2) the unit block is predicted and coded, the intra-frame prediction generates the residual, and the inter-frame prediction performs motion estimation and motion compensation; 3) the residual is transformed, and the transformation coefficient is quantized and scanned; 4) the quantized transformation coefficient, motion information, and prediction information are entropy coded to form a compressed video code stream output; The network abstraction layer mainly defines the data encapsulation format, encapsulates the video data generated by the video encoding layer into data packets of network abstraction layer units, adapts to different network environments and transmits them.

6. The one-way shutter video stream transmission method based on weakly supervised semantic segmentation according to claim 1 is characterized in that: In step S5, the encoding process of Reed-Solomon redundant coding is as follows: 1) Input data vector: The input data is regarded as a vector D=(D1, D2, ..., D n ), there are n data symbols; 2) Coding Matrix: Define the coding matrix B, which is an n×n matrix used to combine the input data with the redundant symbols. The encoded data vector is (D, C) = (D1, D2, …, D n ,C1,C2,…,C m ), where C1, C2, …, C m is a redundant symbol; In matrix operations, the encoding process is expressed as: Y=B×D, where Y is the encoded data, Y includes original data and redundant symbols, D is the original data vector, and B is the encoding matrix.

7. The one-way shutter video stream transmission method based on weakly supervised semantic segmentation according to claim 1 is characterized in that: In step S5, the error correction and data recovery process of Reed-Solomon redundant coding is as follows: 1) Identification of lost data blocks: Assuming that the data vector received by the receiving end is R, some symbols may be lost or erroneous, by comparing the received data vector with the coding matrix, a survivors vector is constructed to represent the correct symbols received; 2) Construct submatrix and recover data: The result after encoding can tolerate at most m data block errors. The encoding matrix corresponding to the remaining correct data blocks is taken out to form an n*n submatrix, denoted as B'. The survivors vector represents the information received by the receiver or indicates that D1, D4, C2 packets are lost. Its corresponding matrix is ​​D. According to the relevant theorem of the matrix, we can find the inverse matrix B' of B' -1 , we can get D = B' -1 *survivors.

Images