Streaming media decoding method and device of network high-definition player and storage medium

Abstract

The invention relates to the technical field of streaming media decoding, and discloses a streaming media decoding method and device for a network high-definition player and a storage medium, and the method comprises the steps: carrying out the multi-thread parallel collection and storage of real-time temperature data, network bandwidth jitter data and video code rate fluctuation data of a decoding chip in the network high-definition player; obtaining a streaming media dynamic feature data set; performing classification training to obtain an inter-frame correlation prediction model; constructing a multi-stage streaming media caching system based on the inter-frame correlation prediction model; monitoring a cache state, a network state and a chip state in the multi-stage streaming media cache system in real time to obtain a decoding event sequence; according to the decoding event sequence, the decoding parameter is adjusted in real time, and the streaming media decoding control instruction is obtained.The collaborative optimization of a network layer, a decoding layer and a display layer is achieved, the playing starting delay is reduced, and the overall stability of a network high-definition player is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of streaming media decoding, and in particular to a method, device and storage medium for decoding streaming media of a network high-definition player. Background Art

[0002] With the popularization of 5G networks and the continuous improvement of video content quality, 4K / 8K ultra-high-definition video streaming services have become an important application scenario for network HD players. However, in the actual decoding process, due to the influence of uncertain factors such as network bandwidth fluctuations, dynamic changes in video bit rates, and heating of decoding chips, it is easy to cause problems such as freezes, mosaics, and asynchrony between audio and video in the playback screen, which seriously affects the user viewing experience.

[0003] Existing decoding technologies mainly use fixed parameter decoding schemes and MMSE-based inter-frame prediction methods. When processing ultra-high-definition video streams of new generation coding standards such as H.265 / HEVC, the complex inter-frame dependencies and limited computing power of playback devices often lead to increased decoding delays. In addition, traditional fixed caching strategies are difficult to cope with network fluctuations and video bit rate changes, and cannot meet the needs of high-quality streaming media playback. Summary of the invention

[0004] The present invention provides a method, device and storage medium for decoding streaming media of a network high-definition player, which realizes the coordinated optimization of the network layer, decoding layer and display layer, reduces the playback startup delay and improves the overall stability of the network high-definition player.

[0005] In a first aspect, the present invention provides a method for decoding streaming media of a network high-definition player, the method comprising: The real-time temperature data of the decoding chip in the network high-definition player, the network bandwidth jitter data, and the video bit rate fluctuation data are collected and stored in parallel through multi-threading to obtain the streaming media dynamic feature data set; Classify and train the data in the streaming media dynamic feature data set according to the inter-frame prediction difficulty to obtain an inter-frame correlation prediction model; Building a multi-level streaming media cache system based on the inter-frame correlation prediction model; Real-time monitoring of the cache status, network status and chip status in the multi-level streaming media cache system to obtain a decoding event sequence; The decoding parameters are adjusted in real time according to the decoding event sequence to obtain streaming media decoding control instructions.

[0006] In a second aspect, the present invention provides a network high-definition player streaming media decoding device, the network high-definition player streaming media decoding device comprising: The parallel acquisition module is used to perform multi-threaded parallel acquisition and storage of real-time temperature data of the decoding chip in the network high-definition player, network bandwidth jitter data, and video bit rate fluctuation data to obtain a streaming media dynamic feature data set; A classification training module, used for classifying and training the data in the streaming media dynamic feature data set according to the inter-frame prediction difficulty to obtain an inter-frame correlation prediction model; A construction module, used for constructing a multi-level streaming media cache system based on the inter-frame correlation prediction model; A real-time monitoring module, used to monitor the cache status, network status and chip status in the multi-level streaming media cache system in real time to obtain a decoding event sequence; The real-time adjustment module is used to adjust the decoding parameters in real time according to the decoding event sequence to obtain the streaming media decoding control instruction.

[0007] A third aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores instructions, which, when executed on a computer, enable the computer to execute the above-mentioned method for decoding streaming media for a network high-definition player.

[0008] In the technical solution provided by the present invention, by constructing a dual-branch neural network structure and combining temporal feature extraction and spatial feature extraction, the decoding prediction accuracy of different types of video frames is significantly improved; based on the hierarchical design of the multi-level streaming media cache system, the coordinated optimization of the network layer, decoding layer and display layer is realized, and the playback startup delay is reduced; the adaptive decoding control strategy of the event trigger mechanism is adopted to reduce the system resource occupancy and the playback jamming rate; the introduction of the attention fusion module improves the accuracy of inter-frame correlation prediction, makes the decoding resource allocation more reasonable, and effectively avoids the problem of overheating of the decoding chip; through multi-task loss function optimization training, the decoding control instructions can take into account temperature control, performance optimization and resource allocation at the same time, which significantly improves the overall stability of the system; based on the dynamic scheduling mechanism of event priority, the system can respond to various abnormal events in a timely manner, ensuring the continuous and stable playback of ultra-high-definition video streams. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0010] Figure 1 A schematic diagram of the steps of a method for decoding streaming media of a network high-definition player according to an embodiment of the present invention; Figure 2It is a structural schematic diagram of a streaming media decoding device for a network high-definition player in an embodiment of the present invention. DETAILED DESCRIPTION

[0011] Embodiments of the present invention provide a method, device and storage medium for decoding streaming media for a network high-definition player. The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0012] For ease of understanding, the specific process of the embodiment of the present invention is described below. Figure 1 , an embodiment of a method for decoding streaming media of a network high-definition player in an embodiment of the present invention includes: Step S1, performing multi-threaded parallel collection and storage of real-time temperature data of a decoding chip in a network high-definition player, network bandwidth jitter data, and video bit rate fluctuation data to obtain a streaming media dynamic feature data set; It is understandable that the execution subject of the present invention may be a network high-definition player streaming media decoding device, or a terminal or a server, which is not limited here. The embodiment of the present invention is described by taking the server as the execution subject as an example.

[0013] Specifically, a data collection mechanism is designed in the player system to ensure that the real-time temperature data of the decoding chip, network bandwidth jitter data, video bit rate fluctuation data and system resource status data are collected and stored in parallel in a multi-threaded manner, so as to construct a streaming media dynamic feature data set. The temperature data of the decoding chip is obtained by periodic sampling, including the measurement of core temperature and surface temperature. The core temperature is directly read from the internal sensor of the decoding chip, while the surface temperature is supplemented by an external temperature sensor to reflect the heat dissipation and thermal stability of the decoding chip. During the data collection process, each sample is accompanied by an accurate timestamp to ensure the accuracy of subsequent data alignment. At the same time, a sliding window detection is performed on the network bandwidth data to obtain the bandwidth jitter. Sliding window detection is a data processing method that continuously calculates the bandwidth change trend within a specified time window and can capture instantaneous fluctuations to obtain a network bandwidth jitter data stream. The data stream includes two key parameters: bandwidth change rate and packet loss rate. The former reflects the dynamic changes in the network transmission rate, while the latter measures the severity of data loss. The two together determine the smoothness of streaming media playback. These parameters are calculated by regularly capturing network traffic packets, and trend prediction is performed in combination with historical data so that appropriate preventive measures can be taken when the network condition shows signs of instability. At the same time, the bit rate of the input video stream is dynamically monitored. Since the bit rate fluctuation of the video stream directly affects the computing load of the decoding chip and the network transmission pressure, the video bit rate is analyzed in real time and its changing trend is recorded. The method based on packet parsing is adopted to extract the bit rate information of each video data from the transmission stream, or directly obtain the video bit rate data before decoding through the decoder module of the player to form a video bit rate fluctuation data stream. The CPU occupancy rate, memory usage rate and GPU usage rate are monitored in real time to form a system resource status data stream. The CPU occupancy rate is monitored through the process management module of the operating system, while the memory usage rate is calculated by analyzing the memory allocation and cache utilization. At the same time, the GPU usage rate is queried through the API interface provided by the graphics card driver. These data together constitute the system resource status data stream, which helps to determine the resource bottleneck problem in the decoding process. The real-time temperature data stream, network bandwidth jitter data stream, video bit rate fluctuation data stream and system resource status data stream are timestamped to ensure that data from different sources are matched according to the same time axis and to construct a multi-dimensional state data matrix. The multi-dimensional state data matrix is ​​stored in a ring buffer to obtain a streaming media dynamic feature data set. A ring buffer is a storage structure that can maintain high access efficiency and avoid memory overflow problems in scenarios where the amount of data is large and needs to be continuously updated. Through ring buffer storage, old data is automatically overwritten by new data, thereby ensuring that the data set is always kept up to date without occupying too many storage resources.

[0014] Step S2, classifying and training the data in the streaming media dynamic feature data set according to the inter-frame prediction difficulty to obtain an inter-frame correlation prediction model; Specifically, the streaming media dynamic feature data set is preprocessed, and the temperature features, network features, and video bit rate features of the decoding chip are standardized to form a temperature feature vector, a network feature vector, and a bit rate feature vector. The temperature feature vector is composed of the core temperature and surface temperature data of the decoding chip, and the mean normalization method is used to make all data within the same scale range. The network feature vector is composed of network bandwidth jitter data and packet loss rate. The sliding window mean filtering method is used to reduce data noise while maintaining the timing characteristics of the data. The bit rate feature vector is composed of video bit rate fluctuation data. The normalization transformation ensures that the bit rate features of different videos are compared within the same numerical range. The data is classified according to the H.265 / HEVC coding standard to distinguish the different characteristics of I frames, P frames, and B frames. I frames are completely independent frames that contain complete image information and do not rely on other frames for decoding. Their coding complexity is usually low, while P frames are predicted based on the previous frame and contain motion vector information. The coding complexity is relatively high. B frames rely on the bidirectional prediction of the previous and next frames, so the motion vector calculation is more complicated. In order to quantify this complexity, the entropy coding complexity and motion vector distribution of each frame type are calculated. The entropy coding complexity is measured by counting the discreteness of each pixel block in the frame, while the motion vector distribution is obtained by analyzing the motion direction and amplitude of the prediction block in the frame. This information constitutes the frame type feature matrix. Based on the frame type feature matrix, a dual-branch neural network structure is constructed, which consists of a temporal feature extraction branch and a spatial feature extraction branch. The temporal feature extraction branch adopts a three-layer bidirectional LSTM network, each layer contains 256 LSTM units. This structure captures the forward and backward temporal dependencies at the same time, so that the network can better understand the dynamic change pattern between video frames. The spatial feature extraction branch adopts the ResNet structure, which contains four residual block groups. Each residual block group consists of three convolutional layers and jump connections. The convolutional layer can extract spatial features within the frame, and the design of the jump connection can alleviate the gradient vanishing problem and improve the training stability of the deep network. The combination of the dual-branch structure enables the model to learn the temporal correlation between frames and extract spatial features within the frame, thereby enhancing the accuracy of decoding prediction. An attention fusion module is added to the output of the dual-branch network. The module consists of a channel attention submodule and a spatial attention submodule. The channel attention submodule increases the focus on key features by calculating the importance weight of each feature channel, while the spatial attention submodule emphasizes the recognition ability of the target area by calculating the weighted distribution of spatial features. After being processed by the attention fusion module, the fused feature map obtained can more effectively express the dynamic change pattern between frames during the decoding process. The fused feature map is input into the decoding prediction head network, which consists of three parallel branches, which are used to predict inter-frame correlation, decoding resource requirements, and cache allocation strategy.Among them, the inter-frame correlation prediction branch estimates the similarity between the current frame and the previous and next frames, thereby guiding the weight adjustment of inter-frame prediction during decoding; the decoding resource demand prediction branch is used to estimate the computing resources required in the decoding process, including CPU occupancy, memory demand, and GPU load, so as to dynamically optimize the decoding parameters; the cache allocation strategy prediction branch calculates a reasonable cache management solution to ensure that there will be no jamming problems due to insufficient or overloaded cache during decoding. All three branches use a two-layer fully connected network for prediction, where the number of neurons in the first fully connected layer is 512, providing sufficient feature expression capabilities, and the number of neurons in the second fully connected layer matches the respective prediction target dimensions to ensure the accuracy of the prediction results. During the training process, a multi-task loss function is used to optimize the decoding prediction head network, where the inter-frame correlation loss is used to measure the predicted inter-frame similarity error, the resource prediction loss is used to evaluate the accuracy of the decoding resource demand prediction, and the cache optimization loss is used to optimize the rationality of the cache management strategy. These loss functions are combined in a weighted manner, and the Adam optimizer is used for gradient update to ensure that the model can converge efficiently and has good generalization ability. After multiple rounds of training and optimization iterations, the model can accurately predict inter-frame correlation, decoding resource requirements, and cache allocation strategies, thereby effectively guiding the decoding process of the streaming media player and improving decoding efficiency and playback stability.

[0015] Step S3, constructing a multi-level streaming media cache system based on the inter-frame correlation prediction model; Specifically, the inter-frame correlation output by the inter-frame correlation prediction model, the decoding resource requirements and the cache allocation strategy are hierarchically mapped to obtain the multi-level cache configuration parameters. The multi-level cache configuration parameters include the network layer cache size, the decoding layer frame type cache ratio and the display layer buffer capacity. These parameters determine the resource allocation method of the cache system at different levels, thereby affecting the stability and efficiency of the entire streaming media decoding. The cache space of the network layer is allocated based on the multi-level cache configuration parameters to form a network layer buffer structure. The size of the network layer buffer structure is dynamically adjusted according to the current video bit rate, and its capacity range is set between two and four times the video bit rate, so as to provide sufficient buffer space when the network fluctuates to reduce the probability of packet loss or freeze. The specific adjustment method of the network layer cache depends on the real-time network bandwidth jitter. When the network is stable, the cache occupancy is reduced to reduce data delay. When the network jitter is large or the packet loss rate is high, the cache capacity is appropriately increased to enhance the anti-jitter capability. At the same time, the data storage strategy of the network layer cache needs to match the data acquisition mechanism of the decoding layer, and adopt an intelligent cache replacement strategy, such as a discard strategy based on timing priority, to ensure that important frames are retained first without affecting the continuity of streaming media playback. After the network layer cache allocation is completed, the decoding layer cache is divided to form a decoding layer cache structure. The cache structure of the decoding layer consists of an I frame cache area, a P frame cache area, and a B frame cache area, and its capacity is dynamically configured according to the analysis results of the inter-frame correlation prediction model. Since the I frame is a key frame for independent decoding, its cache area has the highest priority, so the capacity of the I frame cache area is set to the first target value to ensure that the I frame can be stored and read in time. The P frame depends on the previous frame for decoding, and the capacity of its cache area is set to the second target value, while the B frame is decoded by referring to the previous and next frames at the same time, and its cache demand is relatively low, but in order to prevent decoding anomalies caused by inter-frame prediction errors, the capacity of the B frame cache area is set to the third target value. The division of the decoding layer cache not only needs to consider the cache requirements of different frame types, but also needs to be combined with the prediction results of decoding resource requirements to adjust the cache strategy when the chip resources are limited. For example, when the decoding pressure is high, the capacity of the B frame cache area is reduced to reduce the decoding calculation burden and improve the decoding efficiency. The display layer is cached based on the multi-level cache configuration parameters to ensure that the video frame is smoothly transmitted to the display unit after decoding. The core of the display layer cache structure is two buffers, whose sizes are set to 1.5 times the pixel data required by the current display resolution. This design provides sufficient buffer space when the display frame is switched to avoid screen tearing or frame loss. The display layer cache strategy works in conjunction with the decoding layer cache, that is, when the cache space of the decoding layer is tight, part of the display layer cache is released first to ensure smooth data transmission of the decoding layer. At the same time, when the display frame is stable, the double buffer mechanism is used to seamlessly switch the display frame to reduce screen freezes and improve playback fluency.After completing the configuration of the cache structure of each layer, the data transmission channel between the network layer, decoding layer and display layer is optimized to form an efficient inter-layer data transmission strategy. The design of the inter-layer data transmission strategy takes into account the priority of the data stream. Generally, the transmission priority of I frame data is the highest, followed by P frame, and the transmission priority of B frame is the lowest. Therefore, the data channel dynamically adjusts the transmission scheduling order according to the inter-frame correlation prediction results to ensure the priority transmission of key data during the decoding process. At the same time, the data transmission strategy is combined with the decoding resource status to avoid the buffer overflow caused by insufficient processing power when the decoding layer accumulates data. In terms of implementation, an adaptive flow control mechanism is adopted to dynamically adjust the transmission rate of the data stream by monitoring the length of the decoding queue to ensure that the decoder can run stably. At the same time, the data transmission of the display layer adopts an intelligent scheduling mechanism to reduce frame loss. For example, when the display layer cache is about to be exhausted, the next frame of data is retrieved from the decoding layer cache in advance to maintain smooth playback. By integrating the network layer buffer structure, the decoding layer cache structure, the display layer cache structure and the inter-layer data transmission strategy, a multi-level streaming media cache system is constructed.

[0016] The decoding layer frame type cache ratio and encoding parameters in the multi-level cache configuration parameters are parsed to determine the cache requirements of each frame type. The decoding layer frame type cache ratio reflects the allocation ratio of I frames, P frames and B frames in the cache, while the encoding parameters include the size characteristics and decoding complexity of different frame types under the H.265 / HEVC coding standard. During the parsing process, the analysis results of the inter-frame correlation prediction model are combined to calculate the frame grouping weight values ​​of different frame types. The weight value is used to measure the importance of each frame in the cache and guide the subsequent cache allocation strategy. Since I frames are key frames, they have strong decoding independence and large storage volume, so they are assigned a higher cache weight value. P frames have moderate cache requirements due to their high temporal correlation. B frames rely on previous and next frames for bidirectional prediction. When allocating cache, the storage space is appropriately reduced to optimize the overall cache efficiency. The cache capacity is calculated based on the frame grouping weight value and the total system memory capacity to generate a frame type cache allocation plan. The core of the cache capacity calculation is to ensure that the system memory can be reasonably allocated to different types of frames, while ensuring that the decoding process will not cause data loss or decoding jams due to insufficient cache. During the calculation process, the maximum available memory of the decoding chip, the dynamic memory requirements during system operation, and the bit rate characteristics of the current video stream are comprehensively considered, and dynamic adjustments are made in combination with the historical data provided by the inter-frame correlation prediction model. For example, in the case of high-bitrate video streams, the capacity of the I-frame cache is appropriately increased, while in low-bitrate scenarios, the I-frame cache is reduced and more storage space is reserved for P-frames and B-frames. Through the cache capacity calculation method, the cache space allocation of different frame types is ensured to be reasonable, and the decoding efficiency is maximized within the range allowed by the system memory. The system memory is partitioned and managed based on the frame type cache allocation scheme to generate a cache area physical address table. The partition management of the system memory takes into account the efficiency of data alignment and cache access, so fixed-size memory blocks are used for partitioning. Each partition block corresponds to a specific cache area and is indexed by the physical address table. For example, the I-frame cache area is mapped to the high-priority memory area to ensure its access speed, while the P-frame and B-frame cache areas are mapped to the medium-priority and low-priority memory areas respectively to optimize memory utilization. A multi-level page table structure is constructed based on the cache area physical address table to form a cache access control table. The role of the multi-level page table structure is to provide a more flexible and efficient cache management mechanism, so that the decoding chip can perform fast address conversion when reading cache data and reduce memory fragmentation problems. The multi-level page table structure includes a global page table, a first-level page table, and a second-level page table. The global page table is used to manage the distribution of the entire cache space, the first-level page table is used to map the frame type cache area, and the second-level page table is used to index specific storage blocks. In this way, efficient address lookup is achieved during the decoding process, and the delay of memory access is reduced. At the same time, the dynamic management capability of the cache space is guaranteed, so that the cache space can be reallocated according to the load during system operation.A cache scheduling scheme is designed based on the cache access control table to achieve efficient cache reading and storage. The core of the cache scheduling scheme is to optimize the access order of data and reduce unnecessary cache replacement operations, thereby improving the continuity of decoding. The data access priority of I frames is the highest, so the cache scheduling scheme needs to ensure that the I frame data can be loaded into the decoder as soon as possible and the cache is released in time after decoding to reduce the risk of cache overflow. The cache access scheduling of P frames needs to consider its time correlation with I frames, and load the P frame data to be used in advance through a prediction algorithm, while the cache access scheduling of B frames adopts a delayed loading strategy to reduce the occupation of system memory. In order to optimize the efficiency of cache scheduling, combined with cache prefetching technology, the data that needs to be accessed in the future is predicted by analyzing the playback mode of the video stream, and it is loaded into the cache in advance, thereby reducing the jamming problem caused by the failure of data to arrive in time during the decoding process. A decoding layer cache structure is created based on the cache area physical address table, the cache access control table and the cache scheduling scheme, and the cache structure includes an I frame cache area with a capacity of a first target value, a P frame cache area with a capacity of a second target value, and a B frame cache area with a capacity of a third target value.

[0017] Step S4: real-time monitoring of the cache status, network status and chip status in the multi-level streaming media cache system to obtain a decoding event sequence; Specifically, distributed sampling is performed on the cache layer status in the multi-level streaming media cache system to obtain multi-layer cache status data. Since the cache system involves multiple levels, including network layer cache, decoding layer cache and display layer cache, a distributed sampling strategy is adopted in the sampling process to ensure the integrity and real-time performance of the data. The sampling of the network layer cache status needs to monitor the queue length, data loss rate and buffer utilization of the cache queue, while the sampling of the decoding layer cache status focuses on the cache occupancy and cache hit rate of I frames, P frames and B frames, and the sampling of the display layer cache status tracks the buffer switching rate and screen refresh delay of the frame. Through distributed sampling, high-precision multi-layer cache status data is obtained with low computing overhead. Multi-dimensional threshold analysis is performed on the multi-layer cache status data to identify potential abnormal situations and determine event trigger conditions. Since different video encoding standards and hardware resource configurations will affect the stability of the decoding process, the threshold analysis needs to be dynamically adjusted in combination with video encoding characteristics and hardware resource limitations. A multi-dimensional feature space is established for the cache status data, and the optimal operating range of different parameters is statistically calculated based on historical decoding data to determine a reasonable hierarchical trigger threshold. For example, under the H.265 / HEVC coding standard, the cache hit rate of I frames should be kept within a certain high range, while the cache utilization of P frames and B frames should be adaptively adjusted according to different bit rates. In terms of hardware resources, the temperature of the decoding chip, CPU occupancy rate and GPU load also need to be considered in the threshold analysis to ensure that the system will not affect the decoding quality due to overheating or resource overload. In this way, the event triggering conditions are dynamically determined based on the cache status data, and the threshold setting can be adapted to different video streams and hardware environments. An adaptive priority scheduling mechanism is constructed based on the event triggering conditions, so that dynamic priority is assigned according to the degree of impact of the event on the decoding quality. Since there are many different types of events in the decoding process, such as cache overflow, network jitter, frame loss, and excessive decoding load, a flexible priority scheduling mechanism is established to dynamically adjust the decoding strategy under different circumstances. A weighted scoring method is used to calculate the priority score according to the impact range, duration and historical impact of various events, and a multi-level event priority table is generated. In this process, high-impact events (such as I-frame cache loss or decoder overload) are given higher priority, while low-impact events (such as short-term network fluctuations or slight cache jitter) are arranged at a lower priority, so as to ensure that the system can prioritize key events and avoid affecting the overall decoding performance due to minor events occupying too many computing resources. The multi-level event priority table is rearranged in time sequence and sorted by combining the timestamp information and priority weight of the event to obtain a weighted event sequence. A time-weighted sorting algorithm is adopted, that is, the time factor is considered when calculating the priority of the event, so that new high-priority events are responded to faster, and historical events are appropriately adjusted according to their remaining impact.The weighted event sequence is input into the strategy mapping network to match the response strategy for various events using historical decoding control experience. The strategy mapping network learns the best event response plan from past decoding logs and control strategies based on deep learning or rule matching methods, and provides the best processing strategy for the current event sequence. For example, when the system detects that the network bandwidth drops sharply, the strategy mapping network automatically adjusts the bit rate or increases the cache capacity based on historical experience to reduce playback jams. When the system detects that the temperature of the decoding chip exceeds the safe range, it reduces the decoding frequency or uses hardware acceleration mode to reduce the temperature. The strategy mapping network also needs to have adaptive learning capabilities, that is, it can dynamically optimize the strategy matching scheme in a constantly updated decoding environment to improve the accuracy and efficiency of the response. Based on the weighted event sequence and event processing strategy group, spatiotemporal correlation analysis is performed to identify the temporal dependencies and spatial coupling effects between different events, and obtain the decoding event sequence. The temporal dependency is reflected in the order of occurrence of events and the duration of impact. For example, cache overflow events often lead to the accumulation of decoding queues, which in turn causes frame loss. Therefore, a causal relationship is established in the event sequence, and key events that trigger chain reactions are processed first. The spatial coupling effect is reflected in the mutual influence between different decoding resources. For example, excessive GPU load leads to CPU resource shortage, which affects the execution efficiency of decoding tasks. Therefore, in the process of constructing the event sequence, the impact range of different events is cross-analyzed to identify potential resource competition problems and adjust the decoding scheduling strategy when necessary. By combining timing analysis and spatial correlation analysis, it is ensured that the decoding event sequence can accurately reflect the dynamic evolution of various events in the decoding process.

[0018] Step S5: Adjust the decoding parameters in real time according to the decoding event sequence to obtain streaming media decoding control instructions.

[0019] Specifically, the events in the decoding event sequence are classified and parsed so as to extract corresponding control parameters for different types of events. Decoding events are classified into temperature events, performance events, network events, cache events and bitrate events. Each type of event has different degrees of impact on the decoding process. Therefore, its key features are extracted during the parsing process, and the corresponding decoding control target matrix is ​​established. Temperature events mainly involve the changing trends of the core temperature and surface temperature of the decoding chip. This information is used to adjust the frequency and voltage of the decoding chip to prevent overheating from causing performance degradation or system protection shutdown. Performance events focus on the usage of CPU, GPU and memory, which determines the execution capability and resource scheduling method of the decoding task. Network events involve parameters such as bandwidth jitter, packet loss rate and data delay. This information is crucial for adaptive bitrate adjustment and cache control, while cache events reflect the status of the multi-level cache system, including the cache occupancy rate of I frames, P frames and B frames and the cache overflow. These data are used to optimize the cache management strategy. Bitrate events are mainly related to the dynamic bitrate changes of the input video stream, which affects the decoding complexity and computing resource requirements. Therefore, the decoding strategy needs to be adjusted in a targeted manner. By parsing all these events and constructing a decoding control target matrix, a set of decoding control parameters is formed. Hardware resource control is performed based on the decoding control target matrix to optimize resource allocation and performance management during the decoding process. The frequency and voltage of the decoding chip are adjusted to reduce power consumption at low load and improve computing power at high load, thereby achieving a balance between performance and energy efficiency. The number of decoding threads also needs to be dynamically adjusted to adapt to different decoding task requirements. For example, when playing high-resolution videos, parallel decoding threads are added to increase the decoding speed, while when playing low-bitrate videos, threads are reduced to save computing resources. At the same time, the cache allocation ratio is dynamically adjusted according to the needs of the decoding task. For example, when the network fluctuates greatly, the capacity of the network layer cache is increased to reduce data loss, and when the decoder resources are tight, the occupancy of the B frame cache is reduced to prioritize the storage space of I frames and P frames. Through the above hardware resource control measures, a dynamically optimized hardware control parameter sequence is obtained, which guides the decoding chip to make adaptive adjustments under different conditions. The hardware control parameter sequence is applied to the decoding process of the H.265 / HEVC encoded stream, and the appropriate decoding strategy is matched to ensure that the decoding process can maintain smoothness and optimize the utilization of computing resources. The decoding strategy matching of the H.265 / HEVC encoded stream is mainly based on the decoding priority and frame skipping strategy of I frames, P frames and B frames. Among them, the I frame has the highest priority because it is a key frame and must be fully decoded to ensure the integrity of the basic picture information of the video, while the decoding priority of the P frame is second. When resources are limited, the B frame is skipped to reduce computing overhead.When system resources are tight or network conditions are poor, some B frames are skipped through the intelligent frame skipping strategy, thereby reducing the decoding burden and improving playback fluency. When resources are sufficient, all frames are ensured to be fully decoded to improve picture quality. By combining decoding priority and frame skipping strategy, the playback effect is optimized in different decoding environments to ensure the stability of streaming media playback. After the decoding execution plan is determined, parallel analysis is performed on it to optimize the execution order of decoding tasks and reduce processing delays. The core of parallel analysis is to calculate the dependencies and execution timings between each decoding task, so as to construct a task scheduling sequence, which includes the execution order and time window of the decoding task. During the decoding process, the decoding of I frames needs to be performed first, while the decoding of P frames and B frames is processed in parallel. Therefore, the decoding order needs to be reasonably arranged during task scheduling to make full use of computing resources. For example, a double buffering mechanism is adopted to preload the data of the next frame while decoding the current frame, thereby reducing waiting time and improving decoding efficiency. At the same time, thread parallelism is optimized by task grouping. For example, on a multi-core processor, a high-priority thread is assigned to the decoding of I frames, while multiple threads are used to execute the decoding of P frames and B frames in parallel to maximize the use of hardware resources. By optimizing the dependencies and execution timing of decoding tasks, an efficient task scheduling sequence is generated to improve the overall decoding speed. A resource allocation model is constructed based on the task scheduling sequence, and the optimal resource allocation scheme is obtained by calculating the dynamic balance point of CPU occupancy, memory usage, and GPU usage. During the calculation process, an adaptive load balancing algorithm is used to reasonably allocate decoding tasks among different computing units. For example, when the CPU load is high, some decoding tasks are transferred to the GPU for processing, and when the GPU load is too heavy, the CPU is used for supplementary calculations to avoid overloading of a single computing resource. Memory allocation also needs to be adjusted dynamically to ensure that the decoding cache does not cause data loss or decoding failure due to insufficient memory. By calculating the dynamic balance point of resources, the execution efficiency of decoding tasks is maximized, while the degradation of decoding performance due to insufficient or overloaded resources is avoided. After completing the optimization of the hardware control parameter sequence, decoding execution scheme, and resource allocation scheme, these parameters are integrated and mapped to generate streaming media decoding control instructions. Streaming media decoding control instructions include temperature control instructions, decoding control instructions, and resource control instructions. The temperature control instructions are used to dynamically adjust the power consumption mode of the decoding chip to ensure that the chip operates within a reasonable temperature range. The decoding control instructions are used to adjust the decoding strategy, such as optimizing frame priority or enabling frame skipping mechanism to adapt to different decoding environments, and the resource control instructions are used to dynamically allocate computing resources to ensure that the decoding task can be executed efficiently. By sending these control instructions to the decoding system, precise control of streaming media playback is achieved, so that it can maintain the best playback effect under different network environments and hardware conditions, and effectively improve decoding efficiency and user experience.

[0020] In the embodiment of the present invention, by constructing a dual-branch neural network structure and combining temporal feature extraction and spatial feature extraction, the decoding prediction accuracy of different types of video frames is significantly improved; based on the hierarchical design of the multi-level streaming media cache system, the coordinated optimization of the network layer, decoding layer and display layer is realized, and the playback startup delay is reduced; the adaptive decoding control strategy of the event trigger mechanism is adopted to reduce the system resource occupancy and playback jamming rate; the introduction of the attention fusion module improves the accuracy of inter-frame correlation prediction, makes the decoding resource allocation more reasonable, and effectively avoids the problem of overheating of the decoding chip; through multi-task loss function optimization training, the decoding control instructions can take into account temperature control, performance optimization and resource allocation at the same time, which significantly improves the overall stability of the system; based on the dynamic scheduling mechanism of event priority, the system can respond to various abnormal events in a timely manner, ensuring the continuous and stable playback of ultra-high-definition video streams.

[0021] In a specific embodiment, the process of executing step S1 may specifically include the following steps: Periodically sample the core temperature and surface temperature of the decoding chip in the network HD player to obtain a real-time temperature data stream; Perform sliding window detection on the network bandwidth data to obtain a network bandwidth jitter data stream, which includes a bandwidth change rate and a packet loss rate; Dynamically monitor the bit rate of the input video stream to obtain the video bit rate fluctuation data stream; Collect the status of system resources and obtain the system resource status data stream by monitoring the CPU usage, memory usage, and GPU usage; The real-time temperature data stream, the network bandwidth jitter data stream, the video bit rate fluctuation data stream and the system resource status data stream are timestamped to obtain a multi-dimensional status data matrix; The multi-dimensional state data matrix is ​​stored in a ring buffer to obtain a streaming media dynamic feature data set.

[0022] Specifically, an efficient multi-threaded data acquisition mechanism is established to ensure that all data streams can be acquired in real time and form a complete multi-dimensional state data matrix through timestamp alignment. In terms of decoding chip temperature monitoring, core temperature Represents the temperature inside the decoding chip, while the surface temperature represents the temperature of the decoding chip package surface. The change trend of the two is used to determine the heat dissipation status and workload of the decoding chip. Assume that the sampling interval is , then at any time , the core temperature and the surface temperature are expressed as: in, is the power consumption of the decoding chip, is the heat dissipation capacity, represents the ambient temperature, They represent the influence coefficients of power consumption into temperature change, internal conduction of the chip, package heat transfer, and environmental heat dissipation. These data are periodically sampled to form a real-time temperature data stream. At the same time, network bandwidth jitter detection is an important part of ensuring stable streaming media playback. Set the sliding window length to , then in the window Bandwidth change rate within Calculated as: in, Representative The network bandwidth at the time of sampling is another important indicator of bandwidth jitter data flow, which is the packet loss rate. , and its calculation formula is: in, Representative Window The number of packets lost in Represents the total number of packets in the window. These data reflect changes in network conditions and play an important role in adaptive bitrate adjustment and cache management during decoding. For example, in actual applications, if the bandwidth change rate The packet loss rate is lower than the threshold of 0.05 for a long time. If it exceeds 0.02, it means that the network has significant fluctuations and the video bit rate needs to be appropriately reduced to prevent freezing. For bit rate fluctuation monitoring of input video streams, set the time window Internal bit rate Calculated as: in, Representative time The video data size at Represents the packet transmission time. This data stream is used to evaluate the stability of video transmission in real time. For example, if If the video stream suddenly increases in a short period of time, it means that the video stream has entered a high bit rate segment. At this time, the decoding system needs to reserve more cache space. If the decrease is fast, the cache pre-stored amount is appropriately reduced to save memory resources. At the same time, the system resources are collected. CPU usage , memory usage and GPU usage Obtained through the operating system interface and recorded periodically. Assume that the CPU usage is calculated based on the time of the executed task. and free time ,but: Similarly, memory usage and GPU usage are calculated using similar ratios. The real-time temperature data stream, network bandwidth jitter data stream, video bitrate fluctuation data stream, and system resource status data stream are timestamped to form a multi-dimensional status data matrix. The data sampling interval is set to be consistent with , then at time The data matrix at is represented as: This matrix contains information about the temperature of the decoding chip, network bandwidth jitter, video bit rate changes, and system resource status. In order to efficiently manage this data, a ring buffer storage mechanism is used to ensure that storage overflow will not occur due to the growth of data volume during long-term operation. Assume that the maximum capacity of the ring buffer is , then when the new data When entering the buffer, it is stored as follows: When the buffer is full, new data overwrites the oldest data, so that the most recent data is always kept. The state information of each time step.

[0023] In a specific embodiment, the process of executing step S2 may specifically include the following steps: Preprocessing the streaming media dynamic feature data set to obtain a standardized feature data set, which includes a temperature feature vector, a network feature vector, and a bit rate feature vector; The standardized feature data set is classified according to the I-frame, P-frame and B-frame types in the H.265 / HEVC coding standard, and the entropy coding complexity and motion vector distribution of each frame type are calculated to obtain the frame type feature matrix; A dual-branch neural network structure is constructed based on the frame type feature matrix. The dual-branch neural network structure includes a temporal feature extraction branch and a spatial feature extraction branch. The temporal feature extraction branch is composed of three layers of bidirectional LSTM, and the number of LSTM units in each layer is 256. The spatial feature extraction branch adopts a ResNet structure, which includes four residual block groups, and each residual block group contains three convolutional layers and skip connections. An attention fusion module is added to the output ends of the two branches of the dual-branch neural network structure. The attention fusion module includes a channel attention submodule and a spatial attention submodule. The fused feature map is obtained by calculating the channel weight and spatial weight of the feature map. The fused feature map is input into the decoding prediction head network. The decoding prediction head network contains three parallel branches, which respectively predict inter-frame correlation, decoding resource requirements and cache allocation strategy. Each branch contains two fully connected layers, where the number of neurons in the first layer is 512, and the number of neurons in the second layer matches the prediction target dimension; A multi-task loss function is used to optimize the decoding prediction head network. The multi-task loss function includes inter-frame correlation loss, resource prediction loss and cache optimization loss. The network parameters are iteratively updated through the Adam optimizer to obtain the inter-frame correlation prediction model.

[0024] Specifically, the streaming media dynamic feature dataset is standardized to eliminate the scale differences between different data sources so that it can be efficiently processed by the neural network. In the standardization process, the data needs to be converted into temperature feature vectors, network feature vectors, and bit rate feature vectors, where the temperature feature vector The core temperature of the decoder chip and surface temperature Composition, network feature vector Bandwidth jitter and packet loss rate The bit rate feature vector The instantaneous bit rate of the video stream and frame type bitrate distribution In order to make the numerical range of different features consistent, the mean normalization method is used: in, represents the original data, represents the mean, Represents standard deviation. All features are normalized to a distribution with zero mean and unit variance to improve training stability. The standardized feature dataset is classified according to the H.265 / HEVC coding standard to distinguish the different characteristics of I frames, P frames, and B frames, and the entropy coding complexity and motion vector distribution of each frame type are calculated to construct a frame type feature matrix. Entropy coding complexity It reflects the amount of coded information of the pixel block in the frame. The calculation formula is as follows: in, Is the first The probability distribution of coding units, Represents the number of coding units. The entropy coding complexity of I frames is usually higher, as it needs to store complete picture information, while the entropy coding complexity of P and B frames is relatively low because they are predicted based on reference frames. Motion vector distribution It is an important indicator to measure the complexity of inter-frame prediction, and its calculation method is: in, and Respectively represent The horizontal and vertical components of the motion vector, is the total number of motion vectors in the frame. For B frames, since they rely on the previous and next frames for prediction, they have a more complex motion vector distribution, while the motion vector of P frames mainly depends on the previous frame, and its motion vector distribution is relatively simple. By calculating the entropy coding complexity and motion vector distribution of each frame type, a frame type feature matrix is ​​formed. Based on the frame type feature matrix, a dual-branch neural network structure is constructed, which consists of a temporal feature extraction branch and a spatial feature extraction branch. The temporal feature extraction branch uses a three-layer bidirectional LSTM, and each layer contains 256 LSTM units. The bidirectional LSTM captures the forward and backward dependencies between frames at the same time, and the calculation formula is as follows: in, is the hidden state at the current time step, are input features, and is the weight matrix, is a bias term. The design of bidirectional LSTM makes full use of the temporal correlation between video frames to improve the model's ability to perceive dynamic changes. The spatial feature extraction branch adopts the ResNet structure, which contains four residual block groups. Each residual block group consists of three convolutional layers and jump connections. The calculation formula is: in, represents the input features, represents the weight matrix of the convolutional layer, Represents nonlinear transformation. Residual connection can effectively alleviate the gradient vanishing problem of deep networks and improve training stability. Through the combination of dual-branch structures, temporal features and spatial features are extracted simultaneously to provide more comprehensive information for decoding optimization. At the output end of the dual-branch neural network structure, an attention fusion module is added to improve the expression ability of features. The attention fusion module consists of a channel attention submodule and a spatial attention submodule. The calculation method of the channel attention submodule is: in, Representative channel The weight of is the channel attention weight matrix. The spatial attention submodule is calculated similarly: in, Represents spatial location The weight of is the spatial attention weight matrix. By calculating the channel weights and spatial weights of the feature map, the expressiveness of key features is enhanced and the accuracy of decoding prediction is improved. The fused feature map is input into the decoding prediction head network, which contains three parallel branches, which are used to predict inter-frame correlation, decoding resource requirements and cache allocation strategy respectively. Each branch consists of two fully connected layers, of which the first layer contains 512 neurons for feature transformation, and the calculation formula is: in, is the weight matrix, is the bias term, is a nonlinear activation function. The number of neurons in the second layer matches the prediction target dimension to ensure that the output results meet the decoding optimization requirements. The multi-task loss function is used to optimize the decoding prediction head network. The multi-task loss function includes inter-frame correlation loss, resource prediction loss, and cache optimization loss. The calculation formula is as follows: in, is the inter-frame correlation loss, is the resource forecast loss, is the cache optimization loss, is the loss weight coefficient. The Adam optimizer is used for optimization, and the update rules are as follows: in, represents the model parameters, represents the learning rate, and are the first-order and second-order moment estimates, respectively, It is a numerical stability term. After iterative training, the inter-frame correlation prediction model is finally obtained, which is used to optimize the decoding process of the streaming media player and improve the smoothness and stability of video playback.

[0025] Among them, a multi-task loss function is used to optimize the training of the decoding prediction head network, including: denoising diffusion probability modeling of the three parallel branch outputs of the decoding prediction head network, and obtaining the diffusion probability transfer matrix by constructing a Markov diffusion chain; gradually injecting noise into the prediction results based on the diffusion probability transfer matrix, and obtaining a noisy prediction sequence by setting a forward diffusion process of T time steps; constructing a reverse denoising process for the noisy prediction sequence, and obtaining the noise estimation value for each time step by designing a noise prediction network with a U-Net structure; inputting the noise estimation value into the variational inference network, and minimizing the lower bound of the evidence. Parameter optimization is performed to obtain the posterior distribution parameters; based on the posterior distribution parameters, the output of the decoding prediction head network is probabilistically resampled to generate diversified prediction results and obtain a set of prediction results; the prediction result set is integrated and learned, and multiple sampling results are fused by weighted voting to obtain the final prediction output; the multi-task loss function is calculated based on the predicted output and the true label, and the inter-frame correlation loss, resource prediction loss and cache optimization loss are weightedly combined to obtain the overall loss value; based on the overall loss value, the network parameters are optimized and updated, and the optimized decoding prediction head network is obtained through iterative training with the Adam optimizer.

[0026] In a specific embodiment, the process of executing step S3 may specifically include the following steps: The inter-frame correlation output by the inter-frame correlation prediction model, the decoding resource requirements and the cache allocation strategy are hierarchically mapped to obtain multi-level cache configuration parameters, which include the network layer cache size, the decoding layer frame type cache ratio and the display layer buffer capacity; Based on the multi-level cache configuration parameters, cache space is allocated to the network layer to obtain a network layer buffer structure, and the size of the network layer buffer structure is dynamically adjusted between 2 times and 4 times the video bit rate; Based on the multi-level cache configuration parameters, the decoding layer cache is divided to obtain a decoding layer cache structure, wherein the decoding layer cache structure includes an I frame cache area with a capacity of a first target value, a P frame cache area with a capacity of a second target value, and a B frame cache area with a capacity of a third target value; Based on the multi-level cache configuration parameters, the display layer cache configuration is performed to obtain a display layer cache structure, wherein the display layer cache structure includes two buffers whose sizes are 1.5 times of the pixel data of the current display resolution; Perform data transmission channel configuration on the network layer buffer structure, the decoding layer cache structure and the display layer cache structure to obtain the inter-layer data transmission strategy; The network layer buffer structure, decoding layer cache structure, display layer cache structure and inter-layer data transmission strategy are combined to obtain a multi-level streaming media cache system.

[0027] Specifically, for inter-frame correlation , decoding resource requirements and cache allocation strategy Inter-frame correlation Reflects the degree of redundancy between adjacent frames. When its value is high, it means that the information between frames is relatively similar, and less computing resources are used for prediction compensation during the decoding process; when its value is low, a larger cache and more complex decoding calculations are required. Decoding resource requirements Measure CPU, GPU and memory load, and cache allocation strategy It is used to guide the cache system on how to allocate resources between different layers. Based on these input parameters, multi-level cache configuration parameters are calculated, including the network layer cache size , decoding layer frame type cache ratio and display layer buffer capacity In network layer cache allocation, the size of the network layer buffer structure According to the current video bit rate Dynamic adjustment is performed to ensure the continuity of data transmission. The cache size is set within the range of 2 to 4 times the video bit rate, and its calculation expression is: in, is the dynamic adjustment factor, This is the network buffer compensation time. When the network condition is relatively stable, Take a smaller value to reduce cache usage. When the network jitter is large, Take a larger value to improve data availability and reduce the possibility of jamming. In the decoding layer cache partition, the decoding layer cache structure I-frame buffer , P frame buffer and B-frame buffer The capacity allocation is based on the frame type cache ratio. , the calculation formula is as follows: in, Mainly used to ensure the complete storage of l frame data, Used for storage of P frame prediction data, Since B frames depend on the previous and next frames, their cache requirements are relatively low. The cache ratio is set by the correlation between frames. Impact, when When it is high, the I and P frame buffers are reduced and the E frame buffer is increased. When it is low, the cache ratio of I frames and P frames is increased to enhance the stability of decoding. In the display layer cache configuration, the size of the display layer buffer Depends on the current display resolution and the data storage requirements for each pixel To prevent screen tearing or display delay, two buffers are used, and the size of each buffer should be at least 1.5 times the current display resolution, calculated as follows: in, Represents the number of storage bytes required for each pixel, ensuring that buffer overflow does not occur when data is switched, affecting the display effect. After completing the cache allocation of each layer, the inter-layer data transmission channel is configured to optimize the data transmission efficiency between different layers and obtain the inter-layer data transmission strategy. Set the transmission rate from the network layer to the decoding layer to , the transmission rate from the decoding layer to the display layer is , then it should satisfy: Ensure that there is no transmission bottleneck during data flow. Lower than , resulting in the decoding layer being unable to obtain enough data, causing playback to freeze, so priority must be given to ensuring data transmission in the data transmission strategy. The network layer buffer structure , decoding layer cache structure , Display layer cache structure and inter-layer data transfer strategies The system integrates the cache allocation strategy to form a multi-level streaming media cache system. The system adaptively adjusts the cache allocation strategy under different network environments and decoding conditions to optimize the stability of streaming media playback. When the network bandwidth is low and the jitter is large, the system automatically increases the network layer cache size. , reducing the probability of data loss. When decoding resources are limited, the decoding layer cache division will prioritize the storage of I frames and adjust the cache ratio of P frames to B frames to reduce the computing burden. At the display layer, the double buffering mechanism ensures the smoothness of screen switching and improves the smoothness of video playback. Through the optimization of the multi-level cache system, streaming media playback can maintain a high playback quality under different conditions, while reducing decoding resource consumption and improving playback stability.

[0028] The multi-level streaming media cache system is resource optimized and configured to obtain an optimized multi-level streaming media cache system, including: resource association modeling of the network layer buffer structure and the decoding layer cache structure in the multi-level streaming media cache system, and obtaining a resource utility matrix by calculating the mapping relationship between the cache space and the decoding performance; constructing a multi-objective optimization function based on the resource utility matrix, taking the network bandwidth utilization, the decoding layer cache utilization and the display layer cache switching frequency as optimization targets, and obtaining resource configuration constraints; performing Lagrangian decomposition on the resource configuration constraints, and obtaining a resource allocation objective function by introducing a network cost coefficient and a cache cost coefficient; and designing an iterative optimization function based on the resource allocation objective function. The algorithm is used to calculate the optimal allocation ratio of resources in each cache layer to obtain the frame cache allocation strategy; the I frame cache area, P frame cache area and B frame cache area are dynamically configured according to the frame cache allocation strategy, and the cache optimization scheme is obtained by adjusting the cache space ratio of each frame in real time; the cache optimization scheme is performance evaluated, and the optimization effect evaluation index is obtained by calculating the cache hit rate, access delay and resource utilization; a feedback adjustment mechanism is constructed based on the optimization effect evaluation index, and the optimization parameter update strategy is obtained by dynamically adjusting the resource allocation weight and optimization parameters; the cache optimization scheme and the optimization parameter update strategy are applied to the multi-level streaming media cache system to obtain the optimized multi-level streaming media cache system.

[0029] In a specific embodiment, the execution step divides the cache of the decoding layer based on the multi-level cache configuration parameters to obtain the decoding layer cache structure, which may specifically include the following steps: Parse the decoding layer frame type cache ratio and encoding parameters in the multi-level cache configuration parameters to obtain a frame grouping weight value; Calculate the cache capacity according to the frame grouping weight value and the total system memory capacity to obtain a frame type cache allocation scheme, and partition the system memory based on the frame type cache allocation scheme to obtain a cache area physical address table; Establish a multi-level page table structure for the cache area physical address table to obtain a cache access control table; A cache scheduling scheme is constructed based on a cache access control table, and a decoding layer cache structure is created based on the cache area physical address table, the cache access control table and the cache scheduling scheme. The decoding layer cache structure includes an I frame cache area with a capacity of a first target value, a P frame cache area with a capacity of a second target value, and a B frame cache area with a capacity of a third target value.

[0030] Specifically, the coding structure of the video stream is analyzed, and the cache requirements of different frame types (I frames, P frames, and B frames) are calculated based on the output of the inter-frame correlation prediction model. The cache weight of the I frame is set to , the cache weight of P frame is , the cache weight of B frame is , then according to the H.265 / HEVC coding standard, the frame grouping weight value is calculated as: in, Respectively represent the data size of I frame, P frame and B frame, Respectively represent the average decoding time of the corresponding frame. Since I-frames need to store complete picture information, their data size is relatively large, while P-frames and B-frames rely on other frames for prediction, and their data size is relatively small. However, since B-frames require two frames before and after for bidirectional prediction, their cache requirements are dynamically adjusted according to the specific decoding strategy. By calculating the frame grouping weight value, the priority of each frame type in the decoding layer cache is determined, thereby providing a basis for subsequent cache allocation. After calculating the frame grouping weight value, according to the total system memory capacity Calculate the cache capacity and determine the specific allocation scheme for the frame type cache. Set the total cache capacity of the decoding layer to , then the cache capacity of each frame type is calculated as: in, Represent the cache capacity of I-frame, P-frame and B-frame respectively. Through this allocation method, the cache ratio is dynamically adjusted according to the decoding requirements of different frame types to ensure that high-priority frames (such as I-frames) get enough storage space, while lower-priority frames (such as B-frames) appropriately reduce cache occupancy when resources are limited. After determining the frame type cache allocation scheme, the system memory is partitioned to generate a cache area physical address table. Assume that the minimum storage unit of the memory partition is , then the physical address table is mapped as follows: in, Represents the physical address index of the cache area. Each address index corresponds to a memory block of a fixed size. In order to improve the memory access efficiency, the physical address of the cache blocks of I-frame, P-frame and B-frame is aligned based on continuous address allocation to ensure that the required data can be read quickly during the decoding process. After establishing the cache area physical address table, a multi-level page table structure is constructed to obtain the cache access control table. The page table structure is divided into a global page table and local page tables , where the global page table is used to manage the distribution of the entire cache space, and the local page table is used to index the cache areas of different frame types. The specific mapping method is as follows: in, Represents the local page tables of I-frames, P-frames, and B-frames respectively. Each page table stores the index pointing to the physical address, ensuring that the decoder can quickly locate the corresponding cache area. Based on the cache access control table, a cache scheduling scheme is constructed to optimize data reading and storage strategies. Set cache scheduling priority Affected by inter-frame correlation and cache occupancy The scheduling priority is calculated as: in, and is the adjustment coefficient, The calculation method is: When the cache occupancy When the cache capacity is too high, adjust the cache strategy, such as increasing the I frame cache space to improve decoding stability, or reducing the B frame cache to release memory resources. Combine the cache access control table and the scheduling scheme to form a complete decoding layer cache structure. After completing the cache allocation and scheduling optimization, the decoding layer cache structure includes a capacity of The I frame buffer area has a capacity of The P frame buffer area and capacity are This structure ensures that different types of frame data are stored and accessed according to preset strategies, improving decoding efficiency and stability.

[0031] In a specific embodiment, the process of executing step S4 may specifically include the following steps: Distributed sampling of cache layer states in a multi-level streaming media cache system is performed to obtain multi-layer cache state data; Perform multi-dimensional threshold analysis on multi-layer cache status data, set hierarchical trigger thresholds according to video encoding characteristics and hardware resource limitations, and obtain event trigger conditions; An adaptive priority scheduling mechanism is built based on event trigger conditions, and dynamic priority is assigned according to the impact of events on decoding quality to obtain a multi-level event priority table; The multi-level event priority table is rearranged in time sequence, and the events are sorted according to their timestamp information and priority weights to obtain a weighted event sequence; Input the weighted event sequence into the strategy mapping network, match the response strategy for each type of event based on historical decoding control experience, and obtain the event processing strategy group; Based on the weighted event sequence and event processing strategy group, spatiotemporal correlation analysis is performed to obtain the decoded event sequence by considering the temporal dependency and spatial coupling effect of events.

[0032] Specifically, we build an efficient data collection system to monitor the status of the cache layer in real time. The cache system includes the network layer cache, the decoding layer cache, and the display layer cache. The status of each layer needs to be sampled separately to obtain the complete cache usage. Set the sampling interval of the cache status to , then at any time The cache status is represented as: in, Represents multi-layer cache status data, Represents the network layer cache usage, Represents the decoding layer cache occupancy, Represents the occupancy of the display layer buffer. Through the distributed sampling method, these data are recorded at different time points and the cache status time series is established. After obtaining the cache status data, multi-dimensional threshold analysis is performed to determine whether there are abnormal conditions in the system, and hierarchical trigger thresholds are set according to the video encoding characteristics and hardware resource limitations. The trigger thresholds for each cache layer are set as , when any cache status data exceeds the corresponding threshold, an event can be triggered: in, Represents an event trigger signal. When the value is 1, it means that the current cache state exceeds the threshold and the system needs to take corresponding regulatory measures. The setting of the threshold depends on the video encoding standard and hardware resources. For example, in the H.265 encoding format, the decoding layer cache occupancy of high-bitrate video is usually high, so Need to be increased appropriately, and in a low-bandwidth network environment, Need to be reduced to prevent cache data overload. Based on event trigger conditions, build an adaptive priority scheduling mechanism to dynamically adjust the priority of different events. Set priority value Event Type and decoding quality influencing factors The priority calculation formula is as follows: in, is the adjustment coefficient, The weights representing the event types, for example, a buffer overflow event has a higher weight, while a short-term network fluctuation has a lower weight. Reflects changes in decoding quality. For example, when the frame loss rate is high, Increase, so that the priority of related events is increased. In this way, it is ensured that key events are responded to in a timely manner and the decoding stability of the system is improved. After determining the event priority, the events are rearranged in time sequence and sorted according to the timestamp information and priority weight of the events to obtain a weighted event sequence. Assume that the event list Each event in With timestamp and priority , the event sorting rules are as follows: Among them, sorting is based on timestamp Arrange in descending order, with higher priority events at the front, to ensure that key events can be processed first. After completing the event sorting, the weighted event sequence is input into the strategy mapping network to use historical decoding control experience to match response strategies for various events. Set the strategy mapping function to , then the event handling strategy group is calculated as follows: in, Represents the event handling strategy group, Represents historical decoding control experience. The strategy mapping network uses a deep learning model or a rule-based matching method, such as using a neural network to learn the mapping relationship between historical events and decoding control strategies, or setting predefined strategies based on empirical rules, such as reducing the video bit rate to ensure smooth playback when the network jitter exceeds the threshold. After obtaining the event processing strategy group, a spatiotemporal correlation analysis is performed to comprehensively consider the temporal dependency and spatial coupling effect of the events, and finally generate a decoding event sequence. Set the spatiotemporal correlation function to , then the decoding event sequence The calculation is as follows: in, Represents spatial coupling factors, such as decoding layer buffer overflow causing display layer frame delay, so it is necessary to introduce spatial correlation analysis in the event sequence to optimize the processing order of events. Decoding event sequence It includes different types of events such as abnormal cache status, network jitter, and decoding resource overload, and optimizes the order according to priority and timing. This event sequence is used to adjust streaming media decoding parameters in real time, improve playback stability, reduce the risk of lag, and ensure that the system can provide the best playback experience under different networks and decoding loads. For example, when the system detects that the decoding layer cache is about to overflow, it adjusts the frame drop strategy to reduce the decoding calculation amount, and when the network jitter is severe, it dynamically reduces the bit rate to improve fluency.

[0033] In a specific embodiment, the process of executing step S5 may specifically include the following steps: Classify and analyze the events in the decoding event sequence, extract control parameters based on the features of temperature events, performance events, network events, cache events, and bit rate events, and obtain the decoding control target matrix; Based on the decoding control target matrix, hardware resource control is performed to dynamically configure the frequency and voltage of the decoding chip, the number of decoding threads, and the cache allocation ratio to obtain a hardware control parameter sequence; According to the hardware control parameter sequence, the decoding strategy of H.265 / HEVC coded stream is matched, and the decoding execution scheme is obtained based on the decoding priority and frame skipping strategy of I frame, P frame and B frame; Perform parallel analysis on the decoding execution scheme, and obtain the task scheduling sequence by calculating the dependency and execution timing of each decoding task. The task scheduling sequence includes the execution order and time window of the decoding tasks. Build a resource allocation model based on the task scheduling sequence, and obtain the resource allocation plan by calculating the dynamic balance point of CPU occupancy, memory usage, and GPU usage; The hardware control parameter sequence, the decoding execution plan and the resource allocation plan are integrated and mapped to obtain the streaming media decoding control instructions, which include temperature control instructions, decoding control instructions and resource control instructions.

[0034] Specifically, key features are extracted from the decoding event sequence and classified according to temperature events, performance events, network events, cache events, and bitrate events. Set the decoding event sequence to , it is split into five types of sub-event sets: in, Represents a temperature event, involving the core temperature of the decoded chip and surface temperature Represents a performance event, involving CPU usage , GPU load and memory usage Represents a network event involving bandwidth jitter and packet loss rate Represents cache events, involving cache occupancy and cache overflow rate Represents a bitrate event, involving the instantaneous bitrate of the video stream and bit rate fluctuations By parsing these events and extracting their respective control parameters, a decoding regulation target matrix is ​​constructed: Each row represents a specific decoding control dimension. For example, the first row focuses on hardware temperature and performance, the second row focuses on network and cache status, and the third row focuses on bitrate fluctuations. Based on the decoding control target matrix, adjust the operating parameters of the decoding chip to ensure that the decoding task can be performed in an optimized hardware resource environment. Set the chip frequency to , the voltage is , the number of decoding threads is , the cache allocation ratio is , then the hardware control parameter sequence is calculated as: in, They represent the default chip frequency, voltage, number of decoding threads and cache allocation ratio respectively. is the benchmark threshold set, is the adjustment coefficient. Exceeding the benchmark value When the chip frequency will be lowered to reduce heat, and when the CPU usage Higher than When It will be adjusted appropriately to improve performance, and when the GPU load When too high, the number of decoding threads It will also reduce the cache allocation ratio to reduce GPU pressure. Based on the cache occupancy rate Dynamic adjustments are made to ensure that the cache does not overflow. After obtaining the hardware control parameter sequence, the decoding strategy of the H.265 / HEVC encoded stream is matched, and a decoding execution plan is formulated according to the decoding priority and frame skipping strategy of I frames, P frames, and B frames. Set the l frame priority to , the P frame priority is , B frame priority is , then the decoding priority is calculated as follows: The frame skipping strategy is affected by the decoding resource status Impact, set the frame skip ratio The calculation formula is as follows: when Exceeding the threshold When the system skips B frames first, it will reduce the computing load. After determining the decoding execution plan, perform parallel analysis to optimize the execution order of decoding tasks. Set the task scheduling sequence Decoding tasks are executed in order and time window The parallelism is calculated as follows: when When the number of threads exceeds a certain value, the decoding task needs to reallocate threads to ensure balanced computing load. Based on the task scheduling sequence, a resource allocation model is built and the dynamic balance point of the CPU, GPU, and memory is calculated. Set CPU utilization , GPU utilization , memory usage , then the resource allocation scheme is calculated as follows: if If any of the values ​​exceeds 1, you need to reduce the decoding load, such as adjusting the number of threads. or frame skipping ratio The hardware control parameter sequence, decoding execution plan and resource allocation plan are integrated and mapped to obtain complete streaming media decoding control instructions. Through these control instructions, the system adaptively adjusts the decoding parameters, improves decoding efficiency, reduces freezes, and ensures smooth and stable video playback under different network conditions and computing loads.

[0035] The above describes the method for decoding streaming media of a network high-definition player in the embodiment of the present invention. The following describes the device for decoding streaming media of a network high-definition player in the embodiment of the present invention. Figure 2 In one embodiment of the present invention, a streaming media decoding device for a network high-definition player includes: The parallel acquisition module is used to perform multi-threaded parallel acquisition and storage of real-time temperature data of the decoding chip in the network high-definition player, network bandwidth jitter data, and video bit rate fluctuation data to obtain a streaming media dynamic feature data set; A classification training module is used to classify and train the data in the streaming media dynamic feature data set according to the inter-frame prediction difficulty to obtain an inter-frame correlation prediction model; A construction module, used for constructing a multi-level streaming media cache system based on an inter-frame correlation prediction model; A real-time monitoring module is used to monitor the cache status, network status and chip status in the multi-level streaming media cache system in real time to obtain a decoding event sequence; The real-time adjustment module is used to adjust the decoding parameters in real time according to the decoding event sequence to obtain the streaming media decoding control instruction.

[0036] Through the collaborative cooperation of the above components, by building a dual-branch neural network structure, combined with temporal feature extraction and spatial feature extraction, the decoding prediction accuracy of different types of video frames is significantly improved; based on the hierarchical design of the multi-level streaming media cache system, the coordinated optimization of the network layer, decoding layer and display layer is realized, and the playback startup delay is reduced; the adaptive decoding control strategy based on the event trigger mechanism is adopted to reduce the system resource occupancy and playback jamming rate; the introduction of the attention fusion module improves the accuracy of inter-frame correlation prediction, makes the decoding resource allocation more reasonable, and effectively avoids the problem of overheating of the decoding chip; through multi-task loss function optimization training, the decoding control instructions can take into account temperature control, performance optimization and resource allocation at the same time, which significantly improves the overall stability of the system; the dynamic scheduling mechanism based on event priority enables the system to respond to various abnormal events in a timely manner, ensuring the continuous and stable playback of ultra-high-definition video streams.

[0037] An embodiment of the present invention further provides a computer-readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, the above method is implemented. It can be understood that the computer-readable storage medium in this embodiment can be a volatile readable storage medium or a non-volatile readable storage medium.

[0038] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media provided by the present invention and used in the embodiments may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double-speed data rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM.

[0039] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described systems, systems and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0040] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art or the whole or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program code.

[0041] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features thereof may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for decoding streaming media of a network high-definition player, characterized in that: The method comprises: The real-time temperature data of the decoding chip in the network high-definition player, the network bandwidth jitter data, and the video bit rate fluctuation data are collected and stored in parallel through multi-threading to obtain the streaming media dynamic feature data set; Classify and train the data in the streaming media dynamic feature data set according to the inter-frame prediction difficulty to obtain an inter-frame correlation prediction model; Building a multi-level streaming media cache system based on the inter-frame correlation prediction model; Real-time monitoring of the cache status, network status and chip status in the multi-level streaming media cache system to obtain a decoding event sequence; The decoding parameters are adjusted in real time according to the decoding event sequence to obtain streaming media decoding control instructions.

2. The method for decoding streaming media of a network high-definition player according to claim 1, characterized in that: The real-time temperature data of the decoding chip in the network high-definition player, the network bandwidth jitter data, and the video bit rate fluctuation data are collected and stored in parallel by multi-threading to obtain a streaming media dynamic feature data set, including: Periodically sample the core temperature and surface temperature of the decoding chip in the network HD player to obtain a real-time temperature data stream; Performing sliding window detection on the network bandwidth data to obtain a network bandwidth jitter data stream, wherein the network bandwidth jitter data stream includes a bandwidth change rate and a packet loss rate; Dynamically monitor the bit rate of the input video stream to obtain the video bit rate fluctuation data stream; Collect the status of system resources and obtain the system resource status data stream by monitoring the CPU usage, memory usage, and GPU usage; Performing time stamp alignment on the real-time temperature data stream, the network bandwidth jitter data stream, the video bit rate fluctuation data stream, and the system resource status data stream to obtain a multi-dimensional status data matrix; The multi-dimensional state data matrix is ​​stored in a ring buffer to obtain a streaming media dynamic feature data set.

3. The method for decoding streaming media of a network high-definition player according to claim 1, characterized in that: The data in the streaming media dynamic feature data set is classified and trained according to the inter-frame prediction difficulty to obtain an inter-frame correlation prediction model, including: Performing data preprocessing on the streaming media dynamic feature data set to obtain a standardized feature data set, wherein the standardized feature data set includes a temperature feature vector, a network feature vector, and a bit rate feature vector; The standardized feature data set is classified according to the I frame, P frame and B frame types in the H.265 / HEVC coding standard, and the entropy coding complexity and motion vector distribution of each frame type are calculated to obtain a frame type feature matrix; A dual-branch neural network structure is constructed based on the frame type feature matrix, wherein the dual-branch neural network structure includes a temporal feature extraction branch and a spatial feature extraction branch, wherein the temporal feature extraction branch is composed of a three-layer bidirectional LSTM, and the number of LSTM units in each layer is 256, and the spatial feature extraction branch adopts a ResNet structure, including four residual block groups, and each residual block group contains three convolutional layers and skip connections; Adding an attention fusion module to the two branch output ends of the dual-branch neural network structure, wherein the attention fusion module includes a channel attention submodule and a spatial attention submodule, and obtaining a fused feature map by calculating the channel weight and the spatial weight of the feature map; Input the fused feature map into a decoding prediction head network, wherein the decoding prediction head network comprises three parallel branches for respectively predicting inter-frame correlation, decoding resource requirements, and cache allocation strategy, and each branch comprises two fully connected layers, wherein the number of neurons in the first layer is 512, and the number of neurons in the second layer matches the prediction target dimension; The decoding prediction head network is optimized and trained using a multi-task loss function, where the multi-task loss function includes inter-frame correlation loss, resource prediction loss, and cache optimization loss. The network parameters are iteratively updated through an Adam optimizer to obtain an inter-frame correlation prediction model.

4. The method for decoding streaming media of a network high-definition player according to claim 1, characterized in that: The multi-level streaming media cache system is constructed based on the inter-frame correlation prediction model, including: Performing hierarchical mapping on the inter-frame correlation output by the inter-frame correlation prediction model, the decoding resource requirement and the cache allocation strategy to obtain multi-level cache configuration parameters, wherein the multi-level cache configuration parameters include a network layer cache size, a decoding layer frame type cache ratio and a display layer buffer capacity; Based on the multi-level cache configuration parameters, cache space is allocated to the network layer to obtain a network layer buffer structure, wherein the size of the network layer buffer structure is dynamically adjusted between 2 times and 4 times the video bit rate; Based on the multi-level cache configuration parameters, the decoding layer is cached to obtain a decoding layer cache structure, wherein the decoding layer cache structure includes an I frame cache area with a capacity of a first target value, a P frame cache area with a capacity of a second target value, and a B frame cache area with a capacity of a third target value; Based on the multi-level cache configuration parameters, cache configuration is performed on the display layer to obtain a display layer cache structure, wherein the display layer cache structure includes two buffers whose sizes are 1.5 times of the pixel data of the current display resolution; Performing data transmission channel configuration on the network layer buffer structure, the decoding layer cache structure and the display layer cache structure to obtain an inter-layer data transmission strategy; The network layer buffer structure, the decoding layer cache structure, the display layer cache structure and the inter-layer data transmission strategy are combined to obtain a multi-level streaming media cache system.

5. The method for decoding streaming media of a network high-definition player according to claim 4, characterized in that: The step of dividing the cache of the decoding layer based on the multi-level cache configuration parameters to obtain a decoding layer cache structure includes: Parsing the decoding layer frame type cache ratio and encoding parameters in the multi-level cache configuration parameters to obtain a frame grouping weight value; Calculating the cache capacity according to the frame grouping weight value and the total system memory capacity to obtain a frame type cache allocation scheme, and performing partition management on the system memory based on the frame type cache allocation scheme to obtain a cache area physical address table; Establishing a multi-level page table structure for the cache area physical address table to obtain a cache access control table; A cache scheduling scheme is constructed based on the cache access control table, and a decoding layer cache structure is created based on the cache area physical address table, the cache access control table and the cache scheduling scheme, wherein the decoding layer cache structure includes an I frame cache area with a capacity of a first target value, a P frame cache area with a capacity of a second target value, and a B frame cache area with a capacity of a third target value.

6. The method for decoding streaming media of a network high-definition player according to claim 1, characterized in that: The real-time monitoring of the cache status, network status and chip status in the multi-level streaming media cache system to obtain a decoding event sequence includes: Distributed sampling is performed on cache layer states in the multi-level streaming media cache system to obtain multi-layer cache state data; Performing multi-dimensional threshold analysis on the multi-layer cache status data, setting hierarchical trigger thresholds according to video encoding characteristics and hardware resource limitations, and obtaining event trigger conditions; Based on the event triggering conditions, an adaptive priority scheduling mechanism is constructed, and dynamic priority is assigned according to the impact of the event on the decoding quality to obtain a multi-level event priority table; The multi-level event priority table is rearranged in time sequence, and is sorted according to timestamp information and priority weights of the events to obtain a weighted event sequence; Input the weighted event sequence into the strategy mapping network, match the response strategy for each type of event based on historical decoding control experience, and obtain an event processing strategy group; Based on the weighted event sequence and the event processing strategy group, a spatiotemporal correlation analysis is performed, and a decoded event sequence is obtained by considering the temporal dependency and spatial coupling effect of events.

7. The method for decoding streaming media of a network high-definition player according to claim 1, characterized in that: The real-time adjustment of the decoding parameters according to the decoding event sequence to obtain the streaming media decoding control instruction includes: Classify and analyze the events in the decoding event sequence, extract control parameters according to the features of temperature events, performance events, network events, cache events and bit rate events, and obtain a decoding control target matrix; Based on the decoding control target matrix, hardware resource control is performed to dynamically configure the frequency and voltage of the decoding chip, the number of decoding threads, and the cache allocation ratio to obtain a hardware control parameter sequence; According to the hardware control parameter sequence, a decoding strategy is matched for the H.265 / HEVC coded stream, and a decoding execution scheme is obtained based on the decoding priority and frame skipping strategy of I frames, P frames and B frames; Performing parallel analysis on the decoding execution scheme, and obtaining a task scheduling sequence by calculating the dependency and execution timing of each decoding task, wherein the task scheduling sequence includes the execution order and time window of the decoding tasks; Building a resource allocation model based on the task scheduling sequence, and obtaining a resource allocation plan by calculating the dynamic balance point of CPU occupancy, memory usage, and GPU usage; The hardware control parameter sequence, the decoding execution scheme and the resource allocation scheme are integrated and mapped to obtain streaming media decoding control instructions, which include temperature control instructions, decoding control instructions and resource control instructions.

8. A streaming media decoding device for a network high-definition player, characterized in that: Used to execute the network high-definition player streaming media decoding method according to any one of claims 1 to 7, the network high-definition player streaming media decoding device comprises: The parallel acquisition module is used to perform multi-threaded parallel acquisition and storage of real-time temperature data of the decoding chip in the network high-definition player, network bandwidth jitter data, and video bit rate fluctuation data to obtain a streaming media dynamic feature data set; A classification training module, used for classifying and training the data in the streaming media dynamic feature data set according to the inter-frame prediction difficulty to obtain an inter-frame correlation prediction model; A construction module, used for constructing a multi-level streaming media cache system based on the inter-frame correlation prediction model; A real-time monitoring module, used to monitor the cache status, network status and chip status in the multi-level streaming media cache system in real time to obtain a decoding event sequence; The real-time adjustment module is used to adjust the decoding parameters in real time according to the decoding event sequence to obtain the streaming media decoding control instruction.

9. A computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the processor is enabled to execute the network high-definition player streaming media decoding method as claimed in any one of claims 1 to 7.

Images