Ultra-high-definition video adaptive code rate control method and system based on deep learning

Abstract

The invention discloses an ultra-high-definition video adaptive code rate control method and system based on deep learning, relates to the technical field of video coding transmission, and solves the problem of data congestion or frame loss easily caused by network space fluctuation in the traditional coding behavior. According to the method, the current total cache amount and the to-be-processed video frame flow of the network nodes are comprehensively considered, the characteristic flow is accurately calculated, the network interval which the processing time period needs to reach is further determined, the data volume which can be processed in the processing time period is determined based on the network interval, and the data volume can be processed by analyzing the past data characteristics of the video stream path nodes and other nodes. According to the technical scheme, the interaction flow of the multiple path nodes is accurately controlled, network resources can be reasonably distributed, data congestion caused by too large flow of the network nodes is prevented, it can be ensured that the video code rate reaches the optimal state in the adjustment and verification process, and the video transmission efficiency and quality are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of video coding and transmission, and in particular to a method and system for adaptive bit rate control of ultra-high-definition video based on deep learning. Background Art

[0002] With the rapid development of Internet technology and the widespread adoption of smart devices, ultra-high-definition video has become increasingly common in people's daily lives. For example, 4K, 8K, and even higher-resolution video content is being used in online video playback, video conferencing, virtual reality (VR) and augmented reality (AR). Ultra-high-definition video, with its exquisite image quality, rich details, and lifelike visual effects, provides users with an unprecedented immersive experience and is gradually becoming the mainstream development direction in the video industry.

[0003] Ultra-high-definition video, due to its extremely high resolution and frame rate, results in extremely large amounts of data, placing stringent demands on network transmission bandwidth. In real-world network environments, network conditions are complex and ever-changing, and bandwidth is constantly fluctuating, affected by a variety of factors including network congestion, changes in the number of users, the network service provider's scheduling strategy, and network load variations at different times. For example, during peak network usage periods, when a large number of users are online simultaneously, network bandwidth resources are strained and congestion is prone to occur. Differences in network infrastructure across different regions can also lead to varying bandwidth stability.

[0004] Traditional video bitrate control methods often exhibit significant limitations when applied to data-intensive, real-time applications such as ultra-high-definition video. Some fixed bitrate control strategies cannot dynamically adjust to real-time changes in network bandwidth. When network bandwidth is insufficient, videos frequently freeze, load slowly, or even become unplayable, severely impacting the user experience. Simple adaptive bitrate algorithms, lacking in-depth analysis and accurate prediction of network conditions, struggle to effectively adapt the bitrate of ultra-high-definition video in complex and changing network environments. This inability to fully utilize network bandwidth resources can make it difficult to achieve an ideal balance between video quality and smoothness. Summary of the Invention

[0005] In response to the shortcomings of the existing technology, the present invention provides an ultra-high-definition video adaptive bit rate control method and system based on deep learning, which solves the problem that traditional encoding behavior is easily caused by network space fluctuations, resulting in data congestion or frame loss.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions: a method for adaptive bit rate control of ultra-high-definition video based on deep learning, comprising the following steps:

[0007] Step 1: Identify the network node associated with the ultra-high-definition video. Based on the historical operating data of this network node, identify and record the different bandwidth characteristics associated with different cache capacities within this network node. The specific method is as follows:

[0008] Step 11: Determine different bandwidth values ​​associated with different cache amounts in the cache area from the historical operation data of the network node, and use several groups of bandwidth values ​​associated with a single group of cache amounts as the bandwidth set of the cache amount.

[0009] Step 12: Confirm the bandwidth characteristics of the bandwidth set of a single cache amount: sort the bandwidth values ​​of several groups in the bandwidth set from small to large, and confirm the sorting sequence;

[0010] Randomly select a numerical segment from the sorting sequence, confirm the characteristics of the selected numerical segment, and lock the first bandwidth value D1 of the numerical segment i And the end bandwidth value D2 i , using D2 i -D1 i =Ca i Confirm the numerical range value Ca i , and then the total bandwidth value included in this value segment is calibrated as G i , using Ca i ÷G i =M i , where i represents different numerical segments;

[0011] Will satisfy: M i The numerical segment ≥Y1 is calibrated as the calibrated numerical segment, where Y1 is the preset value;

[0012] Step 13. Select Ca from the confirmed groups of calibration value segments. i The calibration value segment associated with max is recorded as the characteristic value segment. Based on the first bandwidth value and the last bandwidth value associated with the characteristic value segment, the bandwidth feature associated with this cache amount is confirmed. Its bandwidth feature is a characteristic interval. If there is the same Ca i If there are multiple groups of calibrated value segments of max, a group of calibrated value segments is randomly selected as the characteristic value segment and the characteristic interval is confirmed;

[0013] Step 2: Based on the recorded bandwidth characteristics corresponding to different cache sizes, confirm the current video frame traffic volume for ultra-high-definition video to be processed, and based on the total cache size of the network node at the current moment, confirm the network range that needs to be reached during the processing period. The specific method is as follows:

[0014] Step 21. The total cache volume associated with the current network node is calibrated as HH, and the video frame flow rate LH of the ultra-high-definition video to be processed at the current moment is determined by: ZL = LH + HH to determine the characteristic value ZL associated with the current moment. Then, based on the current moment, the unit flow rate DL processed by the corresponding network node in the period between the previous moment and the current moment is determined by: ZL - DL = TL to determine the characteristic flow rate associated with the corresponding network node.

[0015] Step 22: Based on the confirmed characteristic traffic and the different bandwidth characteristics corresponding to different cache amounts, determine the bandwidth characteristics corresponding to this characteristic traffic and mark it as the network range that needs to be reached during the processing period;

[0016] Step 3: Based on the network interval associated with the processing period, determine the amount of data that can be processed during the processing period. Then, based on the past data characteristics of the video stream path node and other nodes, control the flow of multiple path nodes with data interaction. Simultaneously monitor the video frame flow of the video stream path nodes to identify whether there is frame loss during the interaction. The specific method is as follows:

[0017] Step 31: Based on the determined network range, directly lock the data volume range that can be processed within the processing period;

[0018] Step 32: Taking the current moment as the time reference, confirm the data features associated with other path nodes with data interaction in the network node in the past time, and mark the interaction amount associated with other path nodes at the current moment as L k , where k represents different other path nodes;

[0019] For a single group of nodes on other paths: extract the interaction amount from the data features associated with the single group of nodes and satisfy: interaction amount ≤ L k The data features of are selected as features;

[0020] The interaction volume and cache volume associated with different moments are confirmed from the selected features. The following formula is used: interaction volume ÷ (interaction volume + cache volume) = standard feature. The standard features associated with different interaction volumes are confirmed, and the maximum value is selected from several standard features. The interaction volume associated with the maximum value is used as the benchmark volume for a single group of nodes.

[0021] For other path nodes, the same processing method is used to confirm the benchmark quantities associated with the corresponding path nodes in turn;

[0022] Step 33: Based on the baseline data volume determined by other path nodes and the data volume range associated with the processing period, determine the data volume range of the video stream path node within the processing period:

[0023] Sum up several benchmark quantities to determine the total benchmark characteristics of other path nodes and mark them as Zz. Mark the associated data volume interval as [L1, L2]. Use: L1-Zz=SJ1 and L2-Zz=SJ2 to determine the data volume interval [SJ1, SJ2] of the processing period.

[0024] Identify and confirm whether there is a negative value in this data volume interval [SJ1, SJ2]. If so, generate a reference amount reduction signal and execute Step 34 to reconfirm the data volume interval. If not, mark this data volume interval as a determination interval.

[0025] Step 34: Synchronously reduce the baseline quantities confirmed by other path nodes. Each time a set of reduction processes is performed, the baseline quantity is reduced by one unit value. The process stops when there are no negative values ​​in the data column interval corresponding to the total baseline feature associated with several sets of baseline reductions. The currently reduced baseline quantity is recorded, and the corresponding data volume interval is simultaneously marked as the confirmed interval.

[0026] Step 35: During the processing period, the interaction traffic of other path nodes is limited according to the baseline volume recorded by other path nodes. The video frame traffic is interacted through the video stream path node. During the interaction process, the interaction traffic is monitored to identify whether there is frame loss during the interaction process. Based on the identification results, the lost byte data is quickly filled. The specific method is as follows:

[0027] Step 351: Based on the calibrated determination interval, the interactive traffic associated with the video stream path nodes within the processing period is monitored in real time. Interactive traffic that does not fall within the determination interval is identified and recorded as abnormal traffic. The specific time associated with the abnormal traffic is recorded as the abnormal time. The byte data associated with the abnormal time is marked as pending data, and the sequence numbers of the corresponding pending data are confirmed in sequence.

[0028] Step 352: Sort the sequence numbers of the pending data according to the time sequence, confirm the sequence number sequence, and determine whether the adjacent sequence numbers in the sequence number sequence meet the following condition: the next sequence number = the previous sequence number + 1. If not, generate a sequence number missing signal, and lock the missing sequence number based on the sequence number difference between the adjacent sequence numbers. The missing sequence number is determined as follows:

[0029] The next set of adjacent serial numbers is labeled as XL1, and the previous set of serial numbers is labeled as XL2. The missing feature Tz is confirmed using: XL1-XL2=Tz, and then the missing feature Tz is confirmed using: XL2+(Tz-j)=missing serial number, where j is a positive integer and 1≤j<Tz;

[0030] According to the confirmed missing sequence number, the byte data associated with the missing sequence number is filled in real time.

[0031] Preferably, in Step 352, if the adjacent sequence numbers satisfy: the next sequence number = the previous sequence number + 1, then the next group of pending data is calibrated to be correct, and for the pending data ranked first, the sequence number of the byte data at the previous moment of the first group of abnormal moments is selected as the benchmark.

[0032] Preferably, the ultra-high-definition video adaptive bit rate control system based on deep learning includes:

[0033] The bandwidth feature confirmation terminal confirms the network node associated with the ultra-high-definition video and, based on the historical operation data of the network node, confirms and records the different bandwidth features associated with different cache amounts in the network node;

[0034] The network interval confirmation end confirms the current video frame flow of the ultra-high-definition video to be processed based on the recorded bandwidth characteristics corresponding to different cache amounts, and confirms the network interval that needs to be reached during the processing period based on the total cache amount of the network node at the current moment;

[0035] The flow control end determines the amount of data that can be processed in the processing period based on the network interval associated with the processing period. Then, based on the past data characteristics of the video stream path node and other nodes, it controls the flow of multiple path nodes with data interaction.

[0036] The video node monitoring end interacts with the video frame traffic through the video stream path node, and monitors the interactive traffic during the interaction process to identify whether there is frame loss during the interaction process, and quickly fills the lost byte data based on the identification results.

[0037] The present invention provides a method and system for adaptive bit rate control of ultra-high-definition video based on deep learning. Compared with the existing technology, it has the following advantages:

[0038] This invention meticulously analyzes the correlation between cache capacity and bandwidth characteristics in historical network node operation data to determine bandwidth fluctuation characteristic intervals corresponding to different cache capacities. This approach enables more targeted bitrate adjustments based on real-time network conditions during video bitrate regulation, significantly improving the accuracy and effectiveness of bitrate regulation, thereby ensuring smooth playback of ultra-high-definition video in complex network environments.

[0039] Taking into account the current total cache capacity of network nodes and the flow of video frames to be processed, the characteristic flow is accurately calculated, and the network interval that needs to be reached during the processing period is further determined. The amount of data that can be processed during the processing period is determined based on the network interval. By analyzing the past data characteristics of the video stream path node and other nodes, the interactive flow of multiple path nodes is precisely controlled. This not only reasonably allocates network resources and prevents network nodes from experiencing data congestion due to excessive traffic, but also ensures that the video bit rate reaches the optimal state during the adjustment and verification process, thereby improving video transmission efficiency and quality.

[0040] By setting a certain interval to monitor the interactive traffic of the video stream path nodes in real time, abnormal traffic and frame loss can be quickly identified; using serial number sorting and calculation to accurately lock the missing serial numbers and fill the missing byte data in time, it greatly reduces the burden of data interaction monitoring, while ensuring the smoothness of ultra-high-definition video playback, avoiding freezes, and effectively improving the user viewing experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 Schematic diagram of the process of the present invention;

[0042] Figure 2 It is a schematic diagram of the principle framework of the present invention. DETAILED DESCRIPTION

[0043] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0044] First embodiment

[0045] See also Figure 1 , this application provides an ultra-high-definition video adaptive bit rate control method based on deep learning, comprising the following steps:

[0046] Step 1. Confirm the network node associated with this ultra-high-definition video. Based on the historical operation data of this network node, confirm and record the different bandwidth characteristics associated with different cache amounts in this network node. Specifically, the specific network fluctuations of the corresponding network node are generally caused by excessive cache capacity in the cache area, resulting in data congestion and corresponding bandwidth fluctuations. In order to effectively adjust the bit rate of this ultra-high-definition video, it is necessary to determine the fluctuation characteristics of the corresponding bandwidth in advance for effective adjustment;

[0047] The specific method for confirming different bandwidth characteristics associated with different cache characteristics of network nodes is as follows:

[0048] Step 11: Determine different bandwidth values ​​associated with different cache amounts in the cache area from the historical operation data of the network node, and use several groups of bandwidth values ​​associated with a single group of cache amounts as the bandwidth set of the cache amount.

[0049] Step 12: Confirm the bandwidth characteristics of the bandwidth set of a single cache amount: sort the bandwidth values ​​of several groups in the bandwidth set from small to large, and confirm the sorting sequence;

[0050] Randomly select a numerical segment from the sorting sequence, confirm the characteristics of the selected numerical segment, and lock the first bandwidth value D1 of the numerical segment i And the end bandwidth value D2 i , using D2 i -D1 i =Ca i Confirm the numerical range value Ca i , and then the total bandwidth value included in this value segment is calibrated as G i , using Ca i ÷G i =M i , where i represents different numerical segments;

[0051] Will satisfy: M i The numerical segment ≥Y1 is calibrated as the calibration numerical segment, where Y1 is the preset value, and its specific value is determined by the operator based on experience;

[0052] Step 13. Select Ca from the confirmed groups of calibration value segments. i The calibration value segment associated with max is recorded as the characteristic value segment. Based on the first bandwidth value and the last bandwidth value associated with the characteristic value segment, the bandwidth feature associated with this cache amount is confirmed. Its bandwidth feature is a characteristic interval. If there is the same Ca i If there are multiple groups of calibrated value segments of max, a group of calibrated value segments is randomly selected as the characteristic value segment and the characteristic interval is confirmed;

[0053] Specifically, the characteristic interval is the bandwidth fluctuation characteristic corresponding to the corresponding cache amount. In the subsequent processing process, the bandwidth fluctuation characteristic can be used to adjust the video stream bit rate based on the corresponding bandwidth fluctuation characteristic. The bandwidth fluctuation characteristic confirmed by the corresponding cache amount is obvious and has strong characteristics, which is convenient for subsequent feature processing and confirmation.

[0054] Step 2: Based on the recorded bandwidth characteristics corresponding to different cache amounts, confirm the current video frame flow of the ultra-high-definition video to be processed, and based on the total cache amount of the network node at the current moment, confirm the network interval that needs to be reached during the processing period. Specifically, at the current moment, the corresponding network node has a cache process, and its cache process still has relevant data that has not been cached in the past stage. Based on the past data and the current video data that needs to be processed, the network bandwidth is specifically confirmed. The specific method of confirming the network interval is:

[0055] Step 21. The total cache capacity associated with the network node at the current moment is calibrated as HH, and the video frame flow rate LH of the ultra-high-definition video to be processed at the current moment is used to determine the characteristic value ZL associated with the current moment using: ZL = LH + HH. Then, based on the current moment, the unit flow rate DL processed by the corresponding network node between the previous moment and the current moment is determined using: ZL - DL = TL to determine the characteristic flow rate associated with the corresponding network node (that is, the cache capacity that the corresponding network node may generate at the next moment, that is, the specific cache capacity between the current moment and the next moment, which will affect the corresponding bandwidth characteristics);

[0056] Step 22: Based on the confirmed characteristic traffic and the different bandwidth characteristics corresponding to different cache amounts, determine the bandwidth characteristics corresponding to this characteristic traffic and calibrate it as the network range that needs to be reached during the processing period. Specifically, different traffic with different values ​​has different bandwidth characteristics, which will cause different bandwidth fluctuations. In order to confirm the bit rate characteristics associated with different bandwidth conditions, specific adjustments are made;

[0057] Step 3: Based on the network interval associated with the processing period, determine the amount of data that can be processed during the processing period. Then, based on the past data characteristics of the video stream path node and other nodes, control the flow of multiple path nodes with data interaction to ensure that the video bit rate reaches the optimal adjustment and verification state during the actual adjustment and verification process. The specific adjustment method is as follows:

[0058] Step 31: Based on the determined network interval, directly lock the data volume range that can be processed within the processing period (the bandwidth of the network interval is directly related to the corresponding data transmission capacity, so the data volume can be directly confirmed);

[0059] Step 32: Taking the current moment as the time reference, confirm the data features associated with other path nodes with data interaction in the network node in the past time, and mark the interaction amount associated with other path nodes at the current moment as L k , where k represents different other path nodes;

[0060] For a single group of nodes on other paths: extract the interaction amount from the data features associated with the single group of nodes and satisfy: interaction amount ≤ L k The data features of are selected as features;

[0061] The interaction volume and cache volume associated with different moments are confirmed from the selected features. The following formula is used: interaction volume ÷ (interaction volume + cache volume) = standard feature. The standard features associated with different interaction volumes are confirmed, and the maximum value is selected from several standard features. The interaction volume associated with the maximum value is used as the benchmark volume for a single group of nodes.

[0062] For other path nodes, the same processing method is used to confirm the benchmark quantities associated with the corresponding path nodes in turn;

[0063] Step 33: Based on the baseline data volume determined by other path nodes and the data volume range associated with the processing period, determine the data volume range of the video stream path node within the processing period:

[0064] Sum up several benchmark quantities to determine the total benchmark characteristics of other path nodes and mark them as Zz. Mark the associated data volume interval as [L1, L2]. Use: L1-Zz=SJ1 and L2-Zz=SJ2 to determine the data volume interval [SJ1, SJ2] of the processing period.

[0065] Identify and confirm whether there is a negative value in this data volume interval [SJ1, SJ2]. If so, generate a reference amount reduction signal and execute Step 34 to reconfirm the data volume interval. If not, mark this data volume interval as a determination interval.

[0066] Step 34: Synchronously reduce the baseline quantities confirmed by other path nodes. For each set of reduction processes, the baseline quantity is reduced by one unit value (the unit value is a preset value, determined by the operator based on experience, generally 10 bytes). The process stops when there are no negative values ​​in the data column interval corresponding to the total baseline feature associated with several sets of baseline quantity reductions. The currently reduced baseline quantity is recorded, and the corresponding data volume interval is simultaneously marked as the confirmed interval.

[0067] Step 35: During the processing period, the interaction traffic of other path nodes is limited according to the baseline volume recorded by other path nodes (that is, the maximum interaction volume cannot exceed the recorded baseline volume), and the video frame traffic is interacted through the video stream path node. During the interaction process, the interaction traffic is monitored to identify whether there is frame loss during the interaction process (real-time monitoring is performed based on the calibrated determination interval, and the determination interval is a determined benchmark. When the video stream traffic data has frame loss or other problems, it will exceed or fall below this determination interval). The specific method of identification is as follows:

[0068] Step 351: Based on the calibrated determination interval, the interactive traffic associated with the video stream path nodes within the processing period is monitored in real time. Interactive traffic that does not fall within the determination interval is identified and recorded as abnormal traffic. The specific time associated with the abnormal traffic is recorded as the abnormal time. The byte data associated with the abnormal time is marked as pending data, and the sequence numbers of the corresponding pending data are confirmed in sequence (the byte data within the data or each frame is pre-calibrated with a sequence number before sending to facilitate subsequent frame verification).

[0069] Step 352: Sort the sequence numbers of the pending data according to the time sequence, confirm the sequence number sequence, and determine whether the adjacent sequence numbers in the sequence number sequence meet the following condition: next sequence number = previous sequence number + 1. If so, the next set of pending data is correct. For the first set of pending data, the sequence number of the byte data at the moment before the first abnormal moment is selected as the benchmark (that is, among the pending data, the sequence number of the previous set of data is used to assess whether the sequence number of the next set of data meets the standard). If not, a sequence number missing signal is generated, and the missing sequence number is locked based on the sequence number difference between adjacent sequence numbers. The missing sequence number is determined as follows:

[0070] The next set of adjacent serial numbers is labeled as XL1, and the previous set of serial numbers is labeled as XL2. The missing feature Tz is confirmed using: XL1-XL2=Tz, and then the missing feature Tz is confirmed using: XL2+(Tz-j)=missing serial number, where j is a positive integer and 1≤j<Tz;

[0071] Based on the confirmed missing sequence numbers, the byte data associated with the missing sequence numbers are filled in real time to ensure the smoothness of the corresponding ultra-high-definition video during playback without any lag.

[0072] According to the confirmed interval, the monitoring burden in the data interaction process can be effectively reduced. It is only necessary to monitor the corresponding interactive traffic without the need for real-time monitoring of the serial number. Based on the time when the corresponding abnormal traffic exists, the missing byte data can be quickly locked and the missing byte data can be quickly supplemented to achieve a faster and more effective data filling process, which not only ensures the normal playback of ultra-high-definition video, but also achieves a more effective data loss monitoring effect at the same time.

[0073] Second embodiment

[0074] Combine Figure 2 , an ultra-high-definition video adaptive bitrate control system based on deep learning, including:

[0075] The bandwidth feature confirmation terminal confirms the network node associated with the ultra-high-definition video and, based on the historical operation data of the network node, confirms and records the different bandwidth features associated with different cache amounts in the network node;

[0076] The network interval confirmation end confirms the current video frame flow of the ultra-high-definition video to be processed based on the recorded bandwidth characteristics corresponding to different cache amounts, and confirms the network interval that needs to be reached during the processing period based on the total cache amount of the network node at the current moment;

[0077] The flow control end determines the amount of data that can be processed in the processing period based on the network interval associated with the processing period. Then, based on the past data characteristics of the video stream path node and other nodes, it controls the flow of multiple path nodes with data interaction.

[0078] The video node monitoring end interacts with the video frame traffic through the video stream path node, and monitors the interactive traffic during the interaction process to identify whether there is frame loss during the interaction process, and quickly fills the lost byte data based on the identification results.

[0079] Some of the data in the above formulas are dimensionless and numerically calculated. Meanwhile, the contents not described in detail in this specification belong to the prior art known to those skilled in the art.

[0080] The above embodiments are only used to illustrate the technical method of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical method of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical method of the present invention.

Claims

1. An ultra-high-definition video adaptive bit rate control method based on deep learning, characterized in that: The following steps are involved: Step 1: Identify the network node associated with the ultra-high-definition video. Based on the historical operating data of this network node, identify and record the different bandwidth characteristics associated with different cache capacities within this network node. Step 2: Based on the recorded bandwidth characteristics corresponding to different cache sizes, determine the current video frame traffic volume to be processed for ultra-high-definition video, and based on the total cache size of the network node at the current moment, determine the network range that needs to be reached during the processing period; Step 3. Based on the network interval associated with the processing period, confirm the amount of data that can be processed in the processing period. Then, based on the past data characteristics of the video stream path node and other nodes, control the flow of multiple path nodes with data interaction. Simultaneously monitor the video frame flow of the video stream path nodes to identify whether there is frame loss during the interaction process.

2. The method for adaptive bit rate control of ultra-high-definition video based on deep learning according to claim 1, characterized in that: In Step 1, the specific method for confirming different bandwidth characteristics associated with different cache amounts is: Step 11: Determine different bandwidth values ​​associated with different cache amounts in the cache area from the historical operation data of the network node, and use several groups of bandwidth values ​​associated with a single group of cache amounts as the bandwidth set of the cache amount. Step 12: Confirm the bandwidth characteristics of the bandwidth set of a single cache amount: sort the bandwidth values ​​of several groups in the bandwidth set from small to large, and confirm the sorting sequence; Randomly select a numerical segment from the sorting sequence, confirm the characteristics of the selected numerical segment, and lock the first bandwidth value D1 of the numerical segment i And the end bandwidth value D2 i , using D2 i -D1 i =Ca i Confirm the numerical range value Ca i , and then the total bandwidth value included in this value segment is calibrated as G i , using Ca i ÷G i =M i , where i represents different numerical segments; Will satisfy: M i The numerical segment ≥Y1 is calibrated as the calibrated numerical segment, where Y1 is the preset value; Step 13. Select Ca from the confirmed groups of calibration value segments. i The calibrated numerical segment associated with max is recorded as the characteristic value segment. Based on the first bandwidth value and the last bandwidth value associated with the characteristic value segment, the bandwidth feature associated with this cache amount is confirmed, and its bandwidth feature is a characteristic interval.

3. The method for adaptive bit rate control of ultra-high-definition video based on deep learning according to claim 2, characterized in that: In Step 13, if there is the same Ca i If there are multiple groups of calibrated value segments of max, a group of calibrated value segments will be randomly selected as the characteristic value segment and the characteristic interval will be confirmed.

4. The method for adaptive bit rate control of ultra-high-definition video based on deep learning according to claim 1, characterized in that: In Step 2, the specific method of confirming the network range is: Step 21. The total cache volume associated with the current network node is calibrated as HH, and the video frame flow rate LH of the ultra-high-definition video to be processed at the current moment is determined by: ZL = LH + HH to determine the characteristic value ZL associated with the current moment. Then, based on the current moment, the unit flow rate DL processed by the corresponding network node in the period between the previous moment and the current moment is determined by: ZL - DL = TL to determine the characteristic flow rate associated with the corresponding network node. Step 22: Based on the confirmed characteristic traffic and the different bandwidth characteristics corresponding to different cache amounts, confirm the bandwidth characteristics corresponding to this characteristic traffic and mark it as the network range that needs to be reached during the processing period.

5. The method for adaptive bit rate control of ultra-high-definition video based on deep learning according to claim 1, characterized in that: In Step 3, the specific method of performing flow control on multiple path nodes with data interaction is as follows: Step 31: Based on the determined network range, directly lock the data volume range that can be processed within the processing period; Step 32: Taking the current moment as the time reference, confirm the data features associated with other path nodes with data interaction in the network node in the past time, and mark the interaction amount associated with other path nodes at the current moment as L k , where k represents different other path nodes; For a single group of nodes on other paths: extract the interaction amount from the data features associated with the single group of nodes and satisfy: interaction amount ≤ L k The data features of are selected as features; The interaction volume and cache volume associated with different moments are confirmed from the selected features. The following formula is used: interaction volume ÷ (interaction volume + cache volume) = standard feature. The standard features associated with different interaction volumes are confirmed, and the maximum value is selected from several standard features. The interaction volume associated with the maximum value is used as the benchmark volume for a single group of nodes. For other path nodes, the same processing method is used to confirm the benchmark quantities associated with the corresponding path nodes in turn; Step 33: Based on the baseline data volume determined by other path nodes and the data volume range associated with the processing period, determine the data volume range of the video stream path node within the processing period: Sum up several benchmark quantities to determine the total benchmark characteristics of other path nodes and mark them as Zz. Mark the associated data volume interval as [L1, L2]. Use: L1-Zz=SJ1 and L2-Zz=SJ2 to determine the data volume interval [SJ1, SJ2] of the processing period. Identify and confirm whether there is a negative value in this data volume interval [SJ1, SJ2]. If so, generate a reference amount reduction signal and execute Step 34 to reconfirm the data volume interval. If not, mark this data volume interval as a determination interval. Step 34: Synchronously reduce the baseline quantities confirmed by other path nodes. Each time a set of reduction processes is performed, the baseline quantity is reduced by one unit value. The process stops when there are no negative values ​​in the data column interval corresponding to the total baseline feature associated with several sets of baseline reductions. The currently reduced baseline quantity is recorded, and the corresponding data volume interval is simultaneously marked as the confirmed interval. Step 35. During the processing period, the interactive traffic of other path nodes is limited according to the baseline amount recorded by other path nodes, and the video frame traffic is interacted through the video stream path node. During the interaction process, the interactive traffic is monitored to identify whether there is frame loss during the interaction process, and the lost byte data is quickly filled based on the identification results.

6. The method for adaptive bit rate control of ultra-high-definition video based on deep learning according to claim 5, characterized in that: In Step 35, the specific method of identifying whether there is frame loss during the interaction is: Step 351: Based on the calibrated determination interval, the interactive traffic associated with the video stream path nodes within the processing period is monitored in real time. Interactive traffic that does not fall within the determination interval is identified and recorded as abnormal traffic. The specific time associated with the abnormal traffic is recorded as the abnormal time. The byte data associated with the abnormal time is marked as pending data, and the sequence numbers of the corresponding pending data are confirmed in sequence. Step 352: Sort the sequence numbers of the pending data according to the time sequence, confirm the sequence number sequence, and determine whether the adjacent sequence numbers in the sequence number sequence meet the following condition: the next sequence number = the previous sequence number + 1. If not, generate a sequence number missing signal, and lock the missing sequence number based on the sequence number difference between the adjacent sequence numbers. The missing sequence number is determined as follows: The next set of adjacent serial numbers is labeled as XL1, and the previous set of serial numbers is labeled as XL2. The missing feature Tz is confirmed using: XL1-XL2=Tz, and then the missing feature Tz is confirmed using: XL2+(Tz-j)=missing serial number, where j is a positive integer and 1≤j<Tz; According to the confirmed missing sequence number, the byte data associated with the missing sequence number is filled in real time.

7. The method for adaptive bit rate control of ultra-high-definition video based on deep learning according to claim 6, characterized in that: In Step 352, if the adjacent sequence numbers satisfy: the next sequence number = the previous sequence number + 1, then the next group of pending data is calibrated to be correct. For the first pending data, the sequence number of the byte data at the moment before the first abnormal moment is selected as the benchmark.

8. An ultra-high-definition video adaptive bit rate control system based on deep learning, wherein the control system operates according to the ultra-high-definition video adaptive bit rate control method based on deep learning according to any one of claims 1 to 7, characterized in that: include: The bandwidth feature confirmation terminal confirms the network node associated with the ultra-high-definition video and, based on the historical operation data of the network node, confirms and records the different bandwidth features associated with different cache amounts in the network node; The network interval confirmation end confirms the current video frame flow of the ultra-high-definition video to be processed based on the recorded bandwidth characteristics corresponding to different cache amounts, and confirms the network interval that needs to be reached during the processing period based on the total cache amount of the network node at the current moment; The flow control end determines the amount of data that can be processed in the processing period based on the network interval associated with the processing period. Then, based on the past data characteristics of the video stream path node and other nodes, it controls the flow of multiple path nodes with data interaction. The video node monitoring end interacts with the video frame traffic through the video stream path node, and monitors the interactive traffic during the interaction process to identify whether there is frame loss during the interaction process, and quickly fills the lost byte data based on the identification results.

Images