Low-delay adaptive code rate control system for meteorological monitoring video stream

Abstract

The invention discloses a low-delay adaptive code rate control system for a meteorological monitoring video stream, and relates to the technical field of adaptive code rate control in video stream media transmission, and the system comprises a content importance analysis module which is used for carrying out the real-time analysis of a video block to be transmitted, generating a content importance score representing the meteorological information value; and the decision arbitration module is connected with the content importance analysis module and is used for receiving the content importance score, the network state information and the buffer area state information. According to the low-delay adaptive code rate control system for the meteorological monitoring video stream, bandwidth resources can be allocated according to the difference of meteorological information values of video contents, the quality of key monitoring pictures is guaranteed preferentially when network conditions fluctuate, unnecessary code rate switching is inhibited actively through internal look-ahead simulation and risk assessment, and the quality of the meteorological monitoring video stream is improved. Therefore, balance between continuous and clear transmission of high-value information and overall watching fluency is realized in a complex network environment.

Description

A low-latency adaptive bitrate control system for meteorological monitoring video streams Technical Field

[0001] This invention relates to the field of adaptive bitrate control technology in video streaming transmission, specifically a low-latency adaptive bitrate control system for meteorological monitoring video streams. Background Technology

[0002] With the rapid development of meteorological monitoring technology, video streaming has become an important means of meteorological data acquisition, real-time monitoring, and remote analysis. Meteorological monitoring video streams are typically deployed in complex environments such as the field, mountains, and coastlines, requiring high real-time performance, stability, and clarity. Low-latency and high-reliability video transmission are particularly crucial in monitoring severe weather. However, in practical applications, the network environment at meteorological monitoring points is often unstable, with frequent bandwidth fluctuations. Traditional adaptive bitrate (ABR) control technology is prone to causing video stuttering and frequent resolution switching in such scenarios, affecting monitoring effectiveness and the continuity of data analysis.

[0003] Currently, adaptive bitrate control (ABR) technologies are mainly divided into throughput prediction-based, buffer state-based, and hybrid algorithms. With the development of artificial intelligence, reinforcement learning has been introduced into ABR control to improve its decision-making ability in dynamic network environments. Publication number "CN114827683A" describes "A video adaptive bitrate control system and method based on reinforcement learning." This system predicts throughput using an LSTM neural network, combines reinforcement learning for bitrate decision-making, and introduces a smoothing control module to suppress frequent bitrate switching caused by bandwidth fluctuations, thereby optimizing the viewing experience of low-latency video streams.

[0004] The aforementioned patents have improved the adaptability of ABR in dynamic network environments to some extent, but their design is mainly aimed at general video streaming scenarios, such as online video-on-demand and live streaming, and has not fully considered the special needs of meteorological monitoring video streams. Meteorological monitoring video streams typically have the following characteristics: extremely high real-time requirements, with latency needing to be controlled at the millisecond level; the video content contains key meteorological information, such as cloud dynamics and precipitation patterns, and sudden changes in image quality may lead to information loss; they are often located in edge network environments with drastic bandwidth fluctuations, and traditional ABR strategies are prone to causing frequent switching, affecting the continuity of monitoring.

[0005] Therefore, although existing technologies have made improvements in general video stream ABR control, they still fall short of meeting the comprehensive requirements of low latency, high stability, and content consistency in meteorological monitoring scenarios. It is necessary to design a low-latency adaptive bitrate control system for meteorological monitoring video streams, capable of maintaining smooth and real-time video transmission under extreme network fluctuations, ensuring reliable acquisition and analysis of monitoring data. Summary of the Invention

[0006] The purpose of this invention is to provide a low-latency adaptive bitrate control system for meteorological monitoring video streams, so as to solve the problems mentioned in the background art.

[0007] To address the aforementioned technical problems, this invention provides the following technical solution: a low-latency adaptive bitrate control system for meteorological monitoring video streams, comprising: a content importance analysis module, used to perform real-time analysis on video blocks to be transmitted and generate content importance scores characterizing their meteorological information value; a decision arbitration module, connected to the content importance analysis module, used to receive the content importance scores, network status information, and buffer status information, and output a final bitrate decision through an internal multi-level decision mechanism, the multi-level decision mechanism including a preliminary decision unit, a simulated look-ahead verification unit, and a dynamic arbitrator; and an encoding and transmission module, connected to the decision arbitration module, used to encode and transmit the video data according to the final bitrate decision.

[0008] Furthermore, the content importance analysis module includes a lightweight neural network model configured to: extract features from key frames of the video block to be transmitted; identify key content related to meteorological monitoring based on the extracted features; and output a continuously varying value within a predetermined interval as the content importance score, wherein higher values ​​correspond to video frames containing key meteorological change information.

[0009] Furthermore, in the decision arbitration module: the preliminary decision unit is used to generate a preliminary bitrate level suggestion based on the network status information, buffer status information, and content importance score; the simulated look-ahead verification unit, connected to the preliminary decision unit, is used to simulate the transmission process of multiple consecutive video blocks under estimated network conditions with the preliminary bitrate level suggestion as input, and calculate the look-ahead risk index, which is used to assess the potential impact of the preliminary bitrate level suggestion on the transmission quality of subsequent key video blocks and the overall bitrate switching frequency; the dynamic arbitrator, connected to the preliminary decision unit and the simulated look-ahead verification unit, is used to receive the preliminary bitrate level suggestion and the look-ahead risk index, and dynamically adjust the weights of each optimization objective in its internal multi-objective optimization function according to the current content importance score, and output a corrected bitrate decision as the final bitrate decision, wherein the optimization objectives of the multi-objective optimization function include at least keyframe bitrate fidelity and transmission smoothness.

[0010] Furthermore, when the simulated look-ahead verification unit performs the simulation, the estimated network conditions used are obtained through a prediction model based on historical throughput data. The simulation process includes calculating the expected bitrate level of video blocks with content importance scores higher than a set threshold in a predetermined number of video blocks under the simulated transmission path, as well as the number of bitrate switching times in the entire simulation path, and combining the two to calculate the look-ahead risk indicator.

[0011] Furthermore, the dynamic arbitrator dynamically adjusts the weights according to the following rules: when the content importance score is higher than the first threshold, the weight of the keyframe bitrate fidelity optimization target is increased; when the content importance score is lower than the second threshold and the network status information indicates bandwidth fluctuation, the weight of the transmission smoothness optimization target is increased.

[0012] Furthermore, the decision arbitration module also includes: a secondary verification unit connected to the dynamic arbitrator; and a historical policy library storing historical bitrate decision sequences that have been verified to be valid under different network state intervals and content importance score intervals. The secondary verification unit is configured to: compare the corrected bitrate decision output by the dynamic arbitrator with historical decisions in the historical policy library that match the current network state and content importance score; if the deviation of the comparison result exceeds a preset tolerance and the forward-looking risk indicator is higher than a risk threshold, the secondary verification unit triggers policy replacement, selecting the matching historical decision as the final bitrate decision output.

[0013] Furthermore, the network state information includes a future throughput prediction value provided by the network prediction submodule, which uses a recurrent neural network to process the historical throughput sequence to obtain the future throughput prediction value.

[0014] Furthermore, the system also includes a buffer management module located on the client side, which monitors the current full status of the video playback buffer and provides the buffer status information to the decision arbitration module; when the buffer is below the warning level, the buffer management module sends an alarm signal to the decision arbitration module, and when the decision arbitration module receives the alarm signal, it prioritizes the bitrate level that can quickly replenish the buffer.

[0015] Furthermore, the historical bitrate decision sequences in the historical strategy library are derived from stable and valid decision records recorded during the online operation of the system, or from the set of preferred decision strategies generated during the offline training phase of the system; the historical strategy library is updated periodically based on new valid decision records.

[0016] Furthermore, the content importance analysis module, decision arbitration module, and encoding and transmission module are deployed in the edge computing equipment or video acquisition terminal at the meteorological monitoring site; the strategy parameters on which the preliminary decision unit and dynamic arbitrator in the decision arbitration module are based can receive strategy model update data from the remote service center.

[0017] This invention provides a low-latency adaptive bitrate control system for meteorological monitoring video streams. It offers the following advantages: By introducing content-aware and multi-level arbitration decision-making mechanisms, this low-latency adaptive bitrate control system effectively addresses the dual challenges of information fidelity and transmission stability posed by meteorological monitoring video streams. The system can allocate bandwidth resources differentiated based on the meteorological information value of the video content itself, prioritizing the quality of critical monitoring footage during network fluctuations. Simultaneously, through internal forward-looking simulation and risk assessment, it proactively suppresses unnecessary bitrate switching, thereby achieving a balance between continuous and clear transmission of high-value information and overall viewing smoothness in complex network environments.

[0018] This low-latency adaptive bitrate control system for meteorological monitoring video streams features integrated experience-based learning and security design, enhancing its long-term reliability and adaptability in edge environments. A historical policy database and secondary verification mechanism provide a stability safety net based on proven experience, reducing decision-making risks. The architecture, combining edge computing and cloud-based policy collaborative updates, enables the system to continuously optimize and adapt to different monitoring scenarios and network modes while ensuring real-time performance, thereby improving the robustness and effectiveness of the entire meteorological monitoring video data transmission system. Attached Figure Description

[0019] Figure 1 is a data flow diagram between modules of a low-latency adaptive bitrate control system for meteorological monitoring video streams according to the present invention; Figure 2 is a flowchart of the multi-level decision-making mechanism of a low-latency adaptive bitrate control system for meteorological monitoring video streams according to the present invention. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] Please refer to Figures 1 and 2. This invention provides a technical solution: a low-latency adaptive bitrate control system for meteorological monitoring video streams, comprising: a content importance analysis module, used to perform real-time analysis on video blocks to be transmitted and generate content importance scores characterizing their meteorological information value; a decision arbitration module, connected to the content importance analysis module, used to receive content importance scores, network status information, and buffer status information, and output a final bitrate decision through an internal multi-level decision mechanism, the multi-level decision mechanism including a preliminary decision unit, a simulated look-ahead verification unit, and a dynamic arbitrator; and an encoding and transmission module, connected to the decision arbitration module, used to encode and transmit video data according to the final bitrate decision.

[0022] It should be further explained that, during the implementation of this system, the content importance analysis module first processes the next video data block that is about to be transmitted in real time. This module has a built-in lightweight convolutional neural network model trained on meteorological image data. This model performs feature extraction on the key frames of the video block, identifies visual elements directly related to meteorological monitoring value, such as cloud morphology changes, precipitation intensity, and visibility level, and outputs a continuous value between zero and one as the content importance score. This score quantifies the criticality of the meteorological information contained in the current frame.

[0023] The lightweight neural network model employs an improved InceptionNext lightweight architecture, comprising four convolutional layers, one pooling layer, and two fully connected layers. The first two convolutional layers use 3×3 kernels, while the latter two use 1×1 kernels to reduce computational cost. The pooling layer uses max pooling with a 2×2 window size. The fully connected layers have 384 hidden neurons. In the model training parameters, the initial learning rate is set to 0.003, which decays linearly with each training iteration, for a total of 80 iterations. The cross-entropy loss function is used to adapt to the classification requirements. The training dataset used was the ERA5 reanalysis dataset from the European Centre for Medium-Range Weather Forecasts (ECMWF), which contains 66 meteorological variables and covers key meteorological information such as cloud distribution, precipitation areas, and pressure fields. The dataset consists of 800,000 meteorological monitoring video frames. The annotation rules divide the key meteorological information into three levels: Level 1 includes core monitoring targets such as heavy precipitation and severe convection; Level 2 includes important information such as cloud changes and sudden wind speed changes; and Level 3 includes routine information such as clear skies and stable airflows. The annotation process was completed by professional meteorologists in conjunction with meteorological observation data.

[0024] The logic for generating content importance scores is as follows: a lightweight neural network extracts visual features such as texture, brightness, and contours from video frames through convolutional layers, then fuses and maps the extracted features through fully connected layers, finally outputting a continuous numerical score in the range of 0-1. Specifically, video frames containing primary key meteorological information correspond to a score range of 0.7-1.0, those containing secondary key meteorological information correspond to a score range of 0.3-0.7, and those containing only tertiary general information correspond to a score range of 0-0.3. Higher scores indicate richer meteorological monitoring information contained in the video frame, requiring priority in ensuring transmission quality.

[0025] The decision arbitration module, acting as the system's control center, synchronously receives throughput estimates from the network prediction submodule, buffer levels reported by the buffer management module, and the importance scores of the aforementioned content. Its internal workflow begins with the preliminary decision unit, which integrates current status information to generate a preliminary bitrate level suggestion. This suggestion is not directly output but is sent to the simulation look-ahead verification unit for scenario-based analysis. This unit, based on the predicted trend obtained from recent network throughput sequences using a Long Short-Term Memory (LSTM) network model, simulates the transmission of several subsequent video blocks under the estimated network environment if the system adopts the preliminary suggestion, and calculates a look-ahead risk index. This index assesses the potential risk that this transmission path might lead to insufficient bitrate for subsequent high-importance video blocks or excessively frequent overall bitrate switching.

[0026] Subsequently, the dynamic arbitrator simultaneously receives preliminary suggestions and forward-looking risk indicators. Internally, it runs a multi-objective optimization function, which includes at least a fidelity objective to ensure the quality of key images and a smoothness objective to maintain transmission consistency. The dynamic arbitrator dynamically adjusts the weight ratio of these two objectives based on the real-time content importance score. When the content importance score is high, the weight of the fidelity objective increases accordingly, causing the decision to tend to allocate a higher bitrate to key information. When the score is low and network status indicators fluctuate, the weight of the smoothness objective is increased to suppress unnecessary bitrate changes.

[0027] The multi-objective optimization function focuses on keyframe bitrate fidelity and transmission smoothness. Before optimization, the evaluation results of the two objectives are normalized to eliminate the impact of differences in units. Weight adjustment employs a dynamic adaptation mechanism: when the content importance score is higher than 0.7, the weight of keyframe bitrate fidelity is linearly increased, with the slope of the linear increase function determined through calibration using test data from 50 different meteorological scenarios. When network latency fluctuations exceed 50ms, the weight of transmission smoothness is increased through a lookup table mapping relationship. The mapping table pre-sets 10 weight combinations corresponding to different network fluctuation ranges, covering mild, moderate, and severe network fluctuation scenarios. Conflict coordination between objectives is controlled through priority anchor points. When the optimization results of two objectives diverge, priority is given to ensuring the transmission quality of keyframes corresponding to core meteorological information, and then the transmission parameters of non-critical areas are gradually adjusted to balance smoothness.

[0028] The dynamic arbitrator sets its first threshold at 0.8 and its second threshold at 0.3. These values ​​were determined through comparative experiments on 100 different weather scenarios, achieving an optimal balance between decision accuracy and processing efficiency. The first threshold is used to determine the reliability of the initial decision. When the confidence level of the initial decision is higher than 0.8, the decision is directly adopted as the final bitrate strategy. When the confidence level is lower than 0.3, the initial decision is discarded, and the optimal strategy from the historical strategy library is invoked. When the confidence level is between 0.3 and 0.8, the look-ahead verification and dynamic arbitration process are initiated, and the initial decision is corrected by combining look-ahead risk indicators and network prediction results.

[0029] After this weighted multi-objective optimization calculation, the dynamic arbitrator outputs a revised bitrate decision. Before this revised decision takes effect, it needs to be compared with a historical policy library by a secondary verification unit. This historical policy library stores historical decision sequences that have been verified to be stable and effective under different network states and combinations of content importance scores. If the deviation between the revised decision and the matching historical policy exceeds a preset threshold and the forward risk indicator is also high, the system will adopt the corresponding policy from the historical policy library as the final output, thereby ensuring the reliability of the decision in complex and uncertain environments.

[0030] Finally, the encoding and transmission module encodes the video data at the appropriate level and completes the network transmission based on the final bitrate decision determined by the decision arbitration module. This enables the bitrate to be adaptively adjusted according to network conditions and video content value in meteorological monitoring scenarios, taking into account low latency, high stability, and fidelity of key information during transmission.

[0031] The content importance analysis module includes a lightweight neural network model configured to: extract features from key frames of the video block to be transmitted; identify key content related to meteorological monitoring based on the extracted features; and output a continuously varying value within a predetermined interval as a content importance score, where higher values ​​correspond to video frames containing key meteorological change information.

[0032] It should be further explained that in the content importance analysis module, the lightweight neural network model is specifically implemented as a convolutional neural network with optimized and compressed parameters. Its input is the keyframe image data extracted from the video block to be transmitted. This model performs hierarchical feature extraction on the input image through its multiple convolutional and pooling layers. The extracted feature maps are specifically optimized for typical visual patterns in meteorological monitoring scenarios, such as responding to changes in cloud texture density, motion vector features, brightness temperature gradients in precipitation areas, and contrast information related to visibility.

[0033] The neural network model was trained in a supervised manner using a dataset containing a large number of labeled meteorological images. The dataset was labeled based on the knowledge of meteorological experts, and key information such as whether the images contain the initial stage of strong convection, the passage of a front, and the characteristics of severe weather were classified and marked. This enabled the model to learn to map the input visual features to a continuous numerical score.

[0034] The final output layer of the model employs the Sigmoid activation function to ensure that its output content importance score is normalized to a closed interval between zero and one. An output value approaching one indicates that the model determines the image frame contains information of high value for meteorological monitoring and analysis, such as clear structural features of severe weather. Conversely, an output value approaching zero indicates that the image content is relatively static or conventional, with limited information value. This score serves as one of the key input parameters for the subsequent decision arbitration module, providing a direct quantitative basis for the system to implement content-value-based differentiated bitrate control.

[0035] In the decision-making and arbitration module: A preliminary decision-making unit generates preliminary bitrate level recommendations based on network status information, buffer status information, and content importance scores. A simulation look-ahead verification unit, connected to the preliminary decision-making unit, simulates the transmission process of multiple consecutive video blocks under estimated network conditions, using the preliminary bitrate level recommendations as input, and calculates a look-ahead risk index. This index assesses the potential impact of the preliminary bitrate level recommendations on the transmission quality of subsequent key video blocks and the overall bitrate switching frequency. A dynamic arbitrator, connected to the preliminary decision-making unit and the simulation look-ahead verification unit, receives the preliminary bitrate level recommendations and the look-ahead risk index, and dynamically adjusts the weights of each optimization objective in its internal multi-objective optimization function based on the current content importance score, outputting a corrected bitrate decision as the final bitrate decision. The optimization objectives of the multi-objective optimization function include at least keyframe bitrate fidelity and transmission smoothness.

[0036] It should be further explained that, in its specific implementation, the decision arbitration module operates on the following internal data flow and processing logic: The preliminary decision unit receives the future throughput prediction value from the network prediction module, the current buffer status data from the buffer management module, and the content importance score generated by the content importance analysis module; this unit has a built-in decision network trained based on deep reinforcement learning or a policy function with adjustable parameters, which integrates the above input information to generate a preliminary suggested value for the transmission bitrate level of the next video block.

[0037] The initial suggested value was then sent to the simulation look-ahead verification unit, whose core function is to perform internal transmission simulation. Based on a predictive model (such as a Long Short-Term Memory network) with time-series modeling capabilities for network throughput, it constructs a short-term deterministic or probabilistic transmission environment model. Using the initially suggested bitrate level as the starting point, the unit simulates the complete request, transmission, and buffering process for multiple consecutive video blocks. During this process, it tracks two core simulation variables: the actual bitrate level of video blocks whose content importance score exceeds a preset threshold in the simulation path, and the number of bitrate level switches occurring throughout the simulation period. The unit uses a predefined function to fuse these two variables into a comprehensive look-ahead risk indicator value. This indicator quantifies the degree of risk associated with insufficient protection of critical content and a less smooth user experience that may result from adopting the initial suggestion.

[0038] The process of generating forward-looking risk indicators is as follows: First, obtain the content importance scores and preset transmission requirements of the next 5-8 key video blocks following the current video block. Based on the bandwidth prediction results output by the network prediction submodule, evaluate the transmission success rate and expected latency of the subsequent key video blocks under the current bitrate decision. At the same time, calculate the switching probability between the current bitrate and the possible subsequent bitrates, and determine the influence weight of the two by combining the impact of bitrate switching on image quality stability in historical transmission data. Finally, by comprehensively evaluating the transmission quality compliance rate of the subsequent key video blocks and the rationality of bitrate switching, generate forward-looking risk indicators in the range of 0-1. The higher the indicator value, the greater the potential risk of the current bitrate decision.

[0039] The dynamic arbitrator simultaneously receives preliminary bitrate level recommendations and forward-looking risk indicators. Internally, it encapsulates a multi-objective optimization function, which contains at least two mutually constraining optimization objectives: one aims to maximize the bitrate level of high-importance fractional video blocks during transmission (fidelity objective), and the other aims to minimize the sum of the absolute values ​​of the bitrate level differences between adjacent video blocks (smoothness objective).

[0040] The key operation of the dynamic arbitrator lies in its weight adjustment mechanism: this mechanism reads the content importance score of the current video block in real time and dynamically calculates and assigns weight coefficients to the two optimization objectives in this decision based on preset mapping rules related to the score value. When the content importance score increases, the weight coefficient of the fidelity objective increases accordingly, making the optimization solution more inclined to allocate a high bitrate to the current or upcoming keyframes; conversely, when the score is low and other system information suggests network instability, the weight coefficient of the smoothness objective is increased to encourage the stability of the bitrate decision. Finally, the dynamic arbitrator solves this multi-objective optimization problem under the given weight configuration, thereby outputting a corrected bitrate decision that comprehensively considers immediate suggestions, future risks, and content value.

[0041] When the simulated look-ahead verification unit performs the simulation, the estimated network conditions used are obtained through a prediction model based on historical throughput data. The simulation process includes calculating the expected bitrate level of video blocks with content importance scores higher than a set threshold in a predetermined number of video blocks in the simulated transmission path, as well as the number of bitrate switching times in the entire simulation path, and combining the two to calculate a look-ahead risk indicator.

[0042] It should be further explained that when the simulation look-ahead verification unit is running, the estimated network conditions used for simulation come from an independent network prediction submodule. This submodule takes a recent historical throughput data sequence as input, processes it through a long short-term memory network model, and outputs a point estimate or interval estimate sequence of network throughput for several future transmission time slices, which serves as the deterministic or probabilistic bandwidth input required for simulation.

[0043] When the simulation starts, the unit sets the received initial bitrate level suggestion as the decision action for the first simulated video block. Based on the estimated bandwidth sequence, the known video block size, and the current buffer state, it iteratively calculates the transmission process of the subsequent predetermined number of video blocks. In this iterative simulation, the system simultaneously maintains and updates two core tracking variables: the first variable records the bitrate level that the video block whose content importance score in the simulation path exceeds a preset threshold is finally assigned, and the second variable records the cumulative number of times the bitrate level changes between adjacent video blocks during the entire simulation period.

[0044] After the simulation, the unit processes the two tracking variables through a predefined comprehensive evaluation function to calculate the prospective risk index. The design of this comprehensive evaluation function is such that when the expected bitrate level of high-importance video blocks in the simulation path is generally low, or when the number of bitrate switching times in the simulation path is high, the calculated prospective risk index value will change to a higher level, thereby quantitatively reflecting the dual potential risks that may be faced after adopting the initial recommendations.

[0045] The dynamic arbitrator dynamically adjusts the weights according to the following rules: when the content importance score is higher than the first threshold, the weight of the keyframe bitrate fidelity optimization target is increased; when the content importance score is lower than the second threshold and the network status information indicates bandwidth fluctuations, the weight of the transmission smoothness optimization target is increased.

[0046] It should be further explained that the preset dynamic weight adjustment rules within the dynamic arbitrator operate as follows: The system defines two configurable score thresholds, referred to as the first threshold and the second threshold, with the first threshold set higher than the second threshold. When the real-time acquired importance score of the current video block is determined to be higher than the first threshold, the dynamic arbitrator activates the weight adjustment mechanism, increasing the weight coefficient of the keyframe bitrate fidelity objective in the multi-objective optimization function according to predefined rules. This rule can be a linearly increasing function proportional to the portion of the score exceeding the threshold, or a pre-defined lookup table mapping relationship, ensuring that when content is determined to be highly important, the influence of the fidelity objective in the decision-making process is clearly enhanced.

[0047] On the other hand, when the content importance score is determined to be below the second threshold, and indicators parsed from network state information (such as recent throughput variance or specific fluctuation status indicators) confirm that the current network is in a bandwidth fluctuation state, the weight adjustment mechanism shifts to increasing the weight coefficient of the transmission smoothness objective. Under these dual conditions, the weight of the smoothness objective will be strengthened to encourage the optimization function to output the same or adjacent bitrate level as the previous time step, thereby suppressing unproductive bitrate switching that may be caused by network fluctuations during non-critical content transmission.

[0048] The aforementioned weight adjustment process is executed in real time during each decision cycle. The adjusted weight coefficients are directly substituted into the multi-objective optimization function for solution, so that the final output corrected bitrate decision can adaptively respond to the dual changes in content value and network state.

[0049] The decision arbitration module also includes: a secondary verification unit connected to the dynamic arbitrator; and a historical policy library storing historical bitrate decision sequences that have been verified to be valid under different network state intervals and content importance score intervals. The secondary verification unit is configured to: compare the corrected bitrate decision output by the dynamic arbitrator with the historical decisions in the historical policy library that match the current network state and content importance score; if the deviation of the comparison result exceeds the preset tolerance and the forward risk index is higher than the risk threshold, the secondary verification unit triggers policy replacement and selects the matching historical decision as the final bitrate decision output.

[0050] It should be further explained that the specific implementation method of the secondary verification unit working in conjunction with the historical strategy library is as follows: The historical strategy library is a queryable data storage module in structure. Its internal records use network state intervals, content importance score intervals, and corresponding stable bitrate decision sequences as core data items. Among them, the network state interval is defined by discretizing continuous parameters such as throughput prediction value and buffer size into multiple level ranges, and the content importance score interval is also discretized into segments. The bitrate decision sequence stored in each data item comes from the continuous decision records that have been marked as valid and have not caused transmission problems (such as no keyframe quality degradation or excessive stuttering) in the past operation of the system, or from the strategy set optimized for various typical scenarios in the offline simulation environment.

[0051] The matching rules in the historical strategy database are divided into intervals based on network status and content importance score. Network status is divided into three intervals based on bandwidth: low bandwidth (less than 2Mbps), medium bandwidth (2-10Mbps), and high bandwidth (greater than 10Mbps). Content importance score is divided into three levels: 0-0.3, 0.3-0.7, and 0.7-1.0. Each combination of network status interval and content importance score interval corresponds to a set of historically optimal bitrate strategies. Deviation is calculated by comparing the key parameter differences between the current decision strategy and the corresponding combination strategy in the historical strategy database, including bitrate value, adjustment step size, and maintenance duration. The average of these differences is taken as the final deviation. The preset tolerance is set to 0.15, and the risk threshold is set to 0.2. When the deviation is less than the preset tolerance, the historical strategy is directly adopted; when the deviation is greater than the risk threshold, a secondary verification process is triggered.

[0052] When the dynamic arbitrator outputs a corrected bitrate decision, the secondary verification unit is activated. It first performs a query in the historical policy library based on the current real-time network status parameters (such as the level of the current throughput prediction value) and the content importance score of the current video block (the score range to which it belongs), and retrieves one or more historical bitrate decision sequences that match it.

[0053] Subsequently, the secondary verification unit compares the corrected decision output by the dynamic arbitrator with the corresponding decision point in the retrieved historical decision sequence. The core of the comparison is to calculate the decision deviation between the two. This deviation is usually quantified as the absolute value of the difference between the two decision bitrate levels, or as a statistical measure of the difference at multiple consecutive decision points, such as mean square error.

[0054] Simultaneously, the secondary verification unit reads the current value of the forward risk indicator provided by the simulated forward verification unit. The system presets two thresholds: a decision deviation tolerance and a risk threshold. Only when the calculated decision deviation exceeds its tolerance, and the forward risk indicator value also exceeds the risk threshold, does the secondary verification unit determine that the current corrective decision has high uncertainty in terms of both historical experience and future risk, thus triggering a strategy replacement operation. In this operation, the unit abandons the corrective decision and instead selects the corresponding bitrate level from the historical decision sequence retrieved from the historical strategy library that matches the current state as the system's final bitrate decision output.

[0055] If the above conditions are not met simultaneously, the secondary verification unit directly allows the dynamic arbitrator's corrected decision as the final output. The historical policy database has an update mechanism that periodically adds new decision sequences that have been verified to be stable over a long period of time during online operation to the storage, and can eliminate outdated or inefficient records according to preset rules, so that the system's security capabilities can be progressively optimized as operational experience accumulates.

[0056] Network status information includes future throughput predictions provided by the network prediction submodule, which uses a recurrent neural network to process historical throughput sequences to obtain future throughput predictions.

[0057] It should be further explained that, in its implementation, the network prediction submodule continuously monitors and records the historical throughput sequence generated during video block transmission. This sequence is calculated by dividing the size of each transmitted video block by its actual download time and arranged chronologically. The core processing component of this module is a Long Short-Term Memory (LSTM) network model, which is trained to learn the time-dependent patterns of network throughput.

[0058] The network prediction submodule employs a 3-layer LSTM network structure, with 192 hidden units per layer. To avoid overfitting, a dropout layer with a dropout rate of 0.25 is placed between the second and third layers. The length of the model input sequence is determined based on the frame rate of the meteorological video stream. When the video frame rate is 30fps, the input sequence length is set to 45 time steps, with each time step including three types of network state parameters: bandwidth, latency, and packet loss rate. During training, the ratio of the training set to the validation set is set to 8:2. The training data consists of meteorological monitoring network transmission logs from the past 6 months, covering network state data under different time periods and weather conditions. Training is conducted using an iterative batch update method, with each batch containing 64 data sets. The network parameters are adjusted by monitoring the prediction error of the validation set until the error stabilizes within the preset range.

[0059] During operation, the module takes the historical throughput sequence within a recent time window as input and feeds it into the Long Short-Term Memory (LSTM) network. Internally, the network models and extracts the temporal features of the sequence through the coordinated operations of its forget gate, input gate, and output gate, ultimately outputting one or more predicted throughput values ​​for future times. These predicted values ​​undergo necessary post-processing before output, such as inverse normalization to convert them back to actual bit rate units. They are then encapsulated as part of the network state information and provided to the preliminary decision-making unit and the simulation look-ahead verification unit in the decision arbitration module.

[0060] This prediction submodule operates periodically independently of the main decision-making process. Its prediction update frequency matches or is faster than the video block request interval, ensuring that the network state information provided to the decision arbitration module can reflect the latest network change trends and provide key environmental input parameters for forward-looking simulation-based decision-making.

[0061] The system also includes a buffer management module located on the client side, which monitors the current video playback buffer's full status and provides the buffer status information to the decision arbitration module. When the buffer is below the warning level, the buffer management module sends an alarm signal to the decision arbitration module. Upon receiving the alarm signal, the decision arbitration module prioritizes the bitrate level that can quickly replenish the buffer.

[0062] It should be further explained that the buffer management module is deployed on the video playback client and continuously monitors the fullness status of the buffer used to temporarily store downloaded but not yet decoded video data. This status is quantified as the playback duration in the buffer. The module internally has a configurable warning level threshold, which corresponds to a shorter buffer duration.

[0063] The warning level threshold of the buffer management module is set based on the video frame rate and transmission latency characteristics. When the meteorological video stream frame rate is 30fps, the theoretical transmission time of a single frame is 16.6ms. Considering the actual frame processing time in transmission, the warning level threshold is set to 70ms, corresponding to the duration of storing 4 frames in the buffer. The threshold setting fully considers the adaptability to different frame rates. When the frame rate is adjusted to 60fps, the warning level threshold is automatically adjusted to 35ms to ensure that the threshold and frame rate are inversely proportional. At the same time, based on the real-time monitoring results of network transmission latency, when the average latency exceeds 80% of the threshold, the threshold is dynamically increased by 20% to avoid false triggering of frame drops. When the average latency is less than 50% of the threshold, the threshold is decreased by 10% to improve transmission real-time performance.

[0064] During normal operation, this module provides the decision-making arbitration module with the real-time calculated current buffer duration as continuous buffer status information for reference when making routine decisions. When the buffer duration is detected to be decreasing and reaching or falling below a preset warning level threshold, the buffer management module will generate a specific alarm signal, which is different from the regular status data, and immediately send it to the decision-making arbitration module.

[0065] The decision-making arbitration module contains corresponding signal processing logic. When this alarm signal is received, its decision-making process introduces a temporary priority overriding rule: under this rule, the system will temporarily suspend or reduce conventional optimization considerations for objectives such as content importance score, network throughput prediction, and bitrate smoothness. Instead, it will prioritize evaluating and selecting bitrate levels that can fill the buffer at a faster download speed from the available bitrate levels. In other words, it will tend to choose bitrate levels with lower complexity and smaller data volume in order to increase the buffer duration to a safe level in the shortest possible time, thereby avoiding video playback interruption. This mechanism provides the system with a final strategy to ensure playback continuity in special circumstances where buffer resources are scarce.

[0066] The historical bitrate decision sequences in the historical strategy library are derived from stable and valid decision records made during the system's online operation, or from the set of preferred decision strategies generated during the system's offline training phase; the historical strategy library is updated periodically based on new valid decision records.

[0067] It should be further explained that the construction and updating mechanism of the historical strategy database operates as follows: The decision sequences stored in the database come from two sources. The first part comes from the decision history recorded autonomously by the system during actual online operation. When the system continuously executes a series of bitrate decisions and completes the transmission of the corresponding video blocks, if there is no playback stuttering that can be perceived by the user, and the actual bitrate level obtained by the video block that is determined to be of high importance by the content importance analysis module is not lower than the preset protection level during its transmission, and the decision sequence does not trigger the policy replacement operation of the secondary verification unit, then the decision sequence will be marked as stable and valid by the system and can be stored as a candidate record in the historical strategy database.

[0068] The second part originates from the offline training phase. In a controlled simulation environment, by traversing and simulating different typical network bandwidth fluctuation patterns, initial buffer states, and content importance score change curves corresponding to various weather scenarios, reinforcement learning, dynamic programming, or other optimization algorithms are used to solve for each type of scenario and find the decision strategy that performs better after balancing the goals of fidelity and smoothness. These offline-generated strategies constitute a basic set of preferred strategies, which are also imported into the historical strategy library as initial knowledge.

[0069] Regarding the update mechanism, the system initiates the update process periodically (e.g., at fixed time intervals or after processing a fixed number of video blocks). In this process, the system first evaluates new candidate records generated during recent online operation. Only new records that consistently perform stably within a longer observation window and do not cause any transmission anomalies will be officially adopted into the database.

[0070] Meanwhile, the update mechanism also includes elimination logic. Based on the timestamp of the record, the frequency of recent queries and citations, and the evaluation of its decision-making effect under certain changed network environment modes, the existing records in the database are sorted and filtered to remove those that are outdated, inefficient, or have decreased adaptability to the environment, so as to ensure the capacity and timeliness of the strategy database.

[0071] By integrating online experience accumulation with offline prior knowledge and continuously updating itself, the historical strategy library can provide the secondary verification unit with a constantly evolving and reliable decision-making reference that is adapted to the environment.

[0072] The content importance analysis module, decision arbitration module, and encoding and transmission module are deployed in edge computing devices or video acquisition terminals at the meteorological monitoring site; the policy parameters on which the preliminary decision unit and dynamic arbitrator in the decision arbitration module are based can receive policy model update data from the remote service center.

[0073] It should be further explained that the system deployment and update are carried out in the following way: The core functional modules of the entire system, including the content importance analysis module, the decision arbitration module, and the encoding and transmission module, are all integrated and deployed in the edge computing equipment at the meteorological monitoring site or in a dedicated terminal tightly coupled with the video acquisition equipment. This deployment method enables the analysis, decision-making, and encoding of video data to be completed at the network edge close to the data source, effectively reducing the inherent delay caused by data transmission to the remote center for processing, and meeting the real-time requirements of meteorological monitoring.

[0074] Among them, the policy parameters on which the preliminary decision-making unit and the dynamic arbitrator of the decision arbitration module rely, such as the weights of the neural network model, the coefficients in the multi-objective optimization function, and the thresholds in the weight adjustment rules, are not completely static and fixed.

[0075] The system is designed with a secure communication interface, allowing these modules to periodically or on-demand receive optimized and validated policy model update packets from a remote unified policy service center. This update process typically occurs during periods of network idleness or low system load. The update packets are transmitted to edge devices via encrypted channels. After verifying the integrity and signature of the packets, the edge devices securely load the parameters into the corresponding decision units, replacing or merging the original parameter sets. This mechanism enables numerous monitoring terminals deployed across a wide geographical area to receive timely algorithm improvements and utilize the superior models trained by the central server with more comprehensive data. This allows for continuous improvement of the adaptive rate control performance and adaptability of the entire monitoring network without altering the hardware.

[0076] This system effectively addresses the dual challenges of information fidelity and transmission stability posed by meteorological monitoring video streams by introducing content awareness and a multi-level arbitration decision-making mechanism. The system can allocate bandwidth resources differentiated based on the meteorological information value of the video content itself, prioritizing the quality of critical monitoring footage during network fluctuations. Simultaneously, through internal forward-looking simulations and risk assessments, it proactively suppresses unnecessary bitrate switching, thus achieving a balance between continuous and clear transmission of high-value information and overall viewing smoothness in complex network environments.

[0077] Furthermore, the system's integrated experience-learning and security design enhances its long-term reliability and adaptability in edge environments. The historical policy database and secondary verification mechanism provide a stable safety net for decision-making based on proven experience, reducing decision-making risks. The architecture combining edge computing and cloud-based policy collaborative updates enables the system to ensure real-time performance while continuously optimizing and adapting to different monitoring scenarios and network modes, thereby improving the robustness and effectiveness of the entire meteorological monitoring video data transmission system.

[0078] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0079] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A low-latency adaptive bitrate control system for meteorological monitoring video streams, characterized in that, include: The content importance analysis module is used to analyze the video blocks to be transmitted in real time and generate content importance scores that characterize the value of their meteorological information. The decision arbitration module, connected to the content importance analysis module, is used to receive the content importance score, network status information and buffer status information, and output the final bitrate decision through an internal multi-level decision mechanism, which includes a preliminary decision unit, a simulated look-ahead verification unit and a dynamic arbitrator. The encoding and transmission module, connected to the decision arbitration module, is used to encode and transmit video data according to the final bitrate decision.

2. The low-latency adaptive bitrate control system for meteorological monitoring video streams according to claim 1, characterized in that: The content importance analysis module includes a lightweight neural network model configured to: extract features from keyframes of the video block to be transmitted; and identify key content related to meteorological monitoring based on the extracted features. Output a value that changes continuously within a predetermined range as the importance score of the content, where higher values ​​correspond to video frames containing key meteorological change information.

3. A low-latency adaptive bitrate control system for meteorological monitoring video streams according to claim 2, characterized in that: In the decision arbitration module: the preliminary decision unit is used to generate a preliminary bitrate level suggestion based on the network status information, buffer status information, and content importance score; the simulation look-ahead verification unit is connected to the preliminary decision unit and is used to simulate the transmission process of multiple consecutive video blocks under estimated network conditions with the preliminary bitrate level suggestion as input, and calculate the look-ahead risk index. The look-ahead risk index is used to assess the potential impact of the preliminary bitrate level suggestion on the transmission quality of subsequent key video blocks and the overall bitrate switching frequency. The dynamic arbitrator, connected to the preliminary decision-making unit and the simulated look-ahead verification unit, is used to receive the preliminary bitrate level suggestion and the look-ahead risk index, and dynamically adjust the weights of each optimization objective in its internal multi-objective optimization function according to the current content importance score, and output a corrected bitrate decision as the final bitrate decision, wherein the optimization objectives of the multi-objective optimization function include at least keyframe bitrate fidelity and transmission smoothness.

4. A low-latency adaptive bitrate control system for meteorological monitoring video streams according to claim 3, characterized in that: When the simulated look-ahead verification unit performs the simulation, the estimated network conditions used are obtained through a prediction model based on historical throughput data. The simulation process includes calculating the expected bitrate level of video blocks with content importance scores higher than a set threshold in a predetermined number of video blocks in the simulated transmission path, as well as the number of bitrate switching times in the entire simulation path, and combining the two to calculate the look-ahead risk indicator.

5. A low-latency adaptive bitrate control system for meteorological monitoring video streams according to claim 4, characterized in that: The dynamic arbitrator dynamically adjusts the weights according to the following rules: when the content importance score is higher than the first threshold, the weight of the keyframe bitrate fidelity optimization target is increased; when the content importance score is lower than the second threshold and the network status information indicates bandwidth fluctuation, the weight of the transmission smoothness optimization target is increased.

6. A low-latency adaptive bitrate control system for meteorological monitoring video streams according to claim 5, characterized in that: The decision arbitration module further includes: a secondary verification unit connected to the dynamic arbitrator; and a historical policy library storing historical bitrate decision sequences that have been verified to be valid under different network state intervals and content importance score intervals. The secondary verification unit is configured to: compare the corrected bitrate decision output by the dynamic arbitrator with historical decisions in the historical policy library that match the current network state and content importance score; if the deviation of the comparison result exceeds a preset tolerance and the forward-looking risk indicator is higher than a risk threshold, the secondary verification unit triggers policy replacement, selecting the matching historical decision as the final bitrate decision output.

7. A low-latency adaptive bitrate control system for meteorological monitoring video streams according to claim 6, characterized in that: The network state information includes future throughput predictions provided by the network prediction submodule, which uses a recurrent neural network to process historical throughput sequences to obtain the future throughput predictions.

8. A low-latency adaptive bitrate control system for meteorological monitoring video streams according to claim 7, characterized in that: The system also includes a buffer management module located on the client side, which monitors the current video playback buffer's full status and provides the buffer status information to the decision arbitration module. When the buffer is below the warning level, the buffer management module sends an alarm signal to the decision arbitration module. Upon receiving the alarm signal, the decision arbitration module prioritizes selecting the bitrate level that can quickly replenish the buffer.

9. A low-latency adaptive bitrate control system for meteorological monitoring video streams according to claim 8, characterized in that: The historical bitrate decision sequences in the historical strategy library are derived from stable and valid decision records recorded during the online operation of the system, or from the set of preferred decision strategies generated during the offline training phase of the system; the historical strategy library is updated periodically based on new valid decision records.

10. A low-latency adaptive bitrate control system for meteorological monitoring video streams according to claim 9, characterized in that: The content importance analysis module, decision arbitration module, and encoding and transmission module are deployed in the edge computing equipment or video acquisition terminal at the meteorological monitoring site; the strategy parameters on which the preliminary decision unit and dynamic arbitrator in the decision arbitration module are based can receive strategy model update data from the remote service center.

Images