A key frame intelligent extraction task optimization system based on dynamic slice scheduling

Abstract

This invention discloses a keyframe intelligent extraction task optimization system based on dynamic segmentation scheduling, relating to the field of video technology. The key technical points include: collecting historical segmentation scheduling schemes and historical execution effects of keyframe extraction tasks; collecting real-time task feature data of the keyframe extraction task to be processed; constructing a keyframe extraction scheduling graph based on historical segmentation scheduling schemes and historical execution effects, wherein the real-time scheduling factor includes scheduling feature items and scheduling feature values; matching the real-time scheduling factor with the scheduling graph to obtain a target scheduling factor; determining the corresponding target-related scheduling factor set based on the target scheduling factor; and outputting segmentation granularity, scheduling order, and resource allocation strategy according to the target-related scheduling factor set to form the optimal scheduling result for the keyframe intelligent extraction task to be processed, thereby ensuring the accuracy and real-time performance of keyframe extraction.

Description

Technical Field

[0001] This invention relates to the field of video technology, and more specifically, to a keyframe intelligent extraction task optimization system based on dynamic segmentation scheduling. Background Technology

[0002] Keyframe extraction is a fundamental technology in video content analysis, intelligent retrieval, video compression, and security monitoring. Its goal is to extract representative frame sequences that represent the core features of video content from a continuous video stream. With the rapid development of video acquisition technology and multimedia applications, video data exhibits characteristics such as large bitrate fluctuations, high content complexity, and diverse task scenarios. Traditional keyframe extraction task scheduling often adopts fixed segmentation granularity, static scheduling order, and resource allocation strategies, without dynamically adjusting based on the real-time transmission status of the video source, content change characteristics, and the dynamic operation status of computing resources. This easily leads to problems such as overloaded tasks on high-load nodes and idle resources on low-load nodes, resulting in low overall resource utilization and difficulty in ensuring task execution efficiency and stability. Summary of the Invention

[0003] To address the shortcomings of existing technologies, the present invention aims to provide a keyframe intelligent extraction task optimization system based on dynamic fragmentation scheduling.

[0004] To achieve the above objectives, the present invention provides the following technical solution:

[0005] A keyframe intelligent extraction task optimization system based on dynamic fragmentation scheduling includes:

[0006] Acquisition module: Acquires historical segmentation scheduling schemes and historical execution results of keyframe extraction tasks; acquires real-time task feature data of keyframe extraction tasks to be processed;

[0007] Extraction module: Constructs a keyframe extraction scheduling graph based on historical fragmentation scheduling schemes and historical execution effects. The scheduling graph includes scheduling factors and the associated trajectories between scheduling factors. The scheduling factors include factor feature terms and factor feature values.

[0008] Generation module: processes real-time task feature data to generate real-time scheduling factors; wherein, the real-time scheduling factors include scheduling feature items and scheduling feature values;

[0009] Matching module: Matches real-time scheduling factors with the scheduling graph to obtain target scheduling factors; determines the corresponding set of target associated scheduling factors based on the target scheduling factors;

[0010] Output module: Based on the target associated scheduling factor set, output the fragmentation granularity, scheduling order and resource allocation strategy to form the optimal scheduling result of the keyframe intelligent extraction task to be processed.

[0011] Preferably, a keyframe extraction scheduling graph is constructed based on historical fragmentation scheduling schemes and historical execution results, specifically including the following steps:

[0012] By decomposing historical sharding scheduling schemes, relevant sharding scheduling information is obtained; and by filtering historical execution results, effective execution result information is obtained.

[0013] Establish a basic association dataset by establishing a correspondence between information related to segmented scheduling and information on effective execution results;

[0014] A keyframe extraction scheduling graph is constructed based on the type characteristics of the keyframe extraction task and the basic associated dataset.

[0015] Preferably, the information related to the sharding scheduling includes the sharding method, scheduling priority, and resource configuration.

[0016] Preferably, the real-time task feature data is processed to generate a real-time scheduling factor, specifically including the following steps:

[0017] The real-time task feature data includes video source feature data and computing resource feature data;

[0018] Video scheduling factors are obtained based on video source feature data, and resource scheduling factors are obtained based on computing resource feature data.

[0019] Combine video scheduling factors and resource scheduling factors into real-time scheduling factors;

[0020] Feature analysis is performed on real-time scheduling factors to obtain feature analysis results. Based on the feature analysis results and the scheduling factor library, real-time scheduling factors corresponding to the real-time scheduling factors are generated.

[0021] Preferably, obtaining video scheduling factors based on video source feature data specifically includes the following steps:

[0022] Video source regularization feature information is obtained based on video source feature data; wherein, the video source regularization feature information includes effective information on video transmission status and the degree of content change;

[0023] The video state determination result is obtained by performing state determination on the regularized feature information of the video source; wherein, the video state determination result includes the stability of the video transmission process and the frequency of content switching.

[0024] Video scheduling factors that guide task scheduling execution are derived from video status determination results.

[0025] Preferably, obtaining resource scheduling factors based on computational resource characteristic data specifically includes the following steps:

[0026] The resource operation status perception results are obtained by performing operation status perception based on computing resource characteristic information; the resource operation status perception results include the load and availability of computing resources;

[0027] Based on the results of resource operation situation awareness, the support capabilities and constraints of different resources for task execution are determined, and the resource adaptation analysis results are obtained.

[0028] Based on the resource adaptation analysis results, the resource scheduling factors that influence scheduling decisions are identified.

[0029] Preferably, the feature analysis results are obtained by performing feature analysis on the real-time scheduling factors, specifically including the following steps:

[0030] Regularized scheduling factors are obtained by regularizing real-time scheduling factors.

[0031] The correlation determination results of factors are obtained by performing correlation determination on the regularized scheduling factors;

[0032] The factor hierarchy division results based on the factor association discrimination results are used to determine the degree of execution effect;

[0033] Based on the result of factor hierarchy classification, various scheduling factors are extracted to obtain feature analysis results for scheduling matching.

[0034] Preferably, the target scheduling factor is obtained by matching the real-time scheduling factor with the scheduling graph, specifically including the following steps:

[0035] The scheduling feature terms of the real-time scheduling factor are matched with the factor feature terms of the scheduling factor in the scheduling graph to obtain the matching scheduling factor corresponding to the real-time scheduling factor.

[0036] The target scheduling factor corresponding to the real-time scheduling factor is obtained by matching the scheduling feature value of the real-time scheduling factor with the factor feature value of the matching scheduling factor.

[0037] Preferably, determining the corresponding target-associated scheduling factor set based on the target scheduling factor specifically includes the following steps:

[0038] If the real-time scheduling factor corresponding to the target scheduling factor is a video scheduling factor, then obtain the positive pointing mark of the target scheduling factor, obtain the positive correlation trajectory of the target scheduling factor based on the positive pointing mark, and determine all scheduling factors of the positive correlation trajectory as the target positive scheduling factor.

[0039] If the real-time scheduling factor corresponding to the target scheduling factor is a resource scheduling factor, then obtain the reverse pointing mark of the target scheduling factor, obtain the reverse correlation trajectory of the target scheduling factor based on the reverse pointing mark, and determine all scheduling factors in the reverse correlation trajectory as the target reverse scheduling factor.

[0040] The target forward scheduling factor and / or the target reverse scheduling factor constitute the target associated scheduling factor set.

[0041] Preferably, a keyframe extraction scheduling graph is constructed from the basic associated dataset based on the type characteristics of the keyframe extraction task, specifically including the following steps:

[0042] The basic association dataset is classified and organized according to the type characteristics of the keyframe extraction task to obtain the classification association data;

[0043] A keyframe extraction scheduling graph is constructed by combining specific scheduling information and execution effect information of the classified and associated data.

[0044] Compared with the prior art, the present invention has the following beneficial effects:

[0045] This invention acquires historical scheduling data and real-time task features, simultaneously collecting historical sharding scheduling schemes and historical execution effects of keyframe extraction tasks, while also collecting real-time task feature data of the tasks to be processed. A keyframe extraction scheduling graph constructed through the extraction module enables structured and visualized storage of scheduling knowledge. The scheduling graph includes scheduling factors and the correlation trajectories between them, clearly presenting the inherent relationships between different scheduling strategies, scheduling factors, and execution effects. Real-time task feature data is processed to generate standardized real-time scheduling factors, which include scheduling feature items and scheduling feature values, improving the accuracy and versatility of scheduling matching. Based on the target-related scheduling factor set, sharding granularity, scheduling order, and resource allocation strategies are output to generate the optimal scheduling result, achieving efficient, stable, and resource-optimized execution of keyframe extraction tasks. Combining historical scheduling knowledge and real-time status, a suitable sharding granularity is customized for the tasks to be processed, ensuring the accuracy and real-time performance of keyframe extraction. Attached Figure Description

[0046] Figure 1 This invention provides a schematic diagram of a keyframe intelligent extraction task optimization system based on dynamic fragmentation scheduling, as an embodiment of the present invention.

[0047] Figure 2 This is a schematic diagram illustrating the steps of obtaining a real-time scheduling factor in a keyframe intelligent extraction task optimization system based on dynamic fragmentation scheduling, as provided in an embodiment of the present invention. Detailed Implementation

[0048] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0049] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0050] Secondly, the term "an embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single embodiment or an embodiment selectively excluded from other embodiments.

[0051] Reference Figures 1-2 As shown.

[0052] The embodiments further illustrate the keyframe intelligent extraction task optimization system based on dynamic fragmentation scheduling proposed in this invention.

[0053] A keyframe intelligent extraction task optimization system based on dynamic fragmentation scheduling includes:

[0054] Acquisition module: Acquires historical segmentation scheduling schemes and historical execution results of keyframe extraction tasks; acquires real-time task feature data of keyframe extraction tasks to be processed;

[0055] Extraction module: Based on historical sharding scheduling schemes and historical execution results, a keyframe extraction scheduling graph is constructed. The scheduling graph includes scheduling factors and the associated trajectories between scheduling factors. The scheduling factors include factor feature terms and factor feature values.

[0056] Generation module: processes real-time task feature data to generate real-time scheduling factors; wherein, the real-time scheduling factors include scheduling feature items and scheduling feature values;

[0057] Matching module: Matches real-time scheduling factors with the scheduling graph to obtain target scheduling factors; determines the corresponding set of target associated scheduling factors based on the target scheduling factors;

[0058] Output module: Based on the target associated scheduling factor set, output the fragmentation granularity, scheduling order and resource allocation strategy to form the optimal scheduling result of the keyframe intelligent extraction task to be processed.

[0059] Based on historical scheduling knowledge and execution data carried by the target-related scheduling factor set, a comprehensive optimal scheduling scheme adapted to the tasks to be processed is generated to achieve a match between business needs and resource capabilities. The target-related scheduling factor set integrates the positive correlation scheduling factors on the video service side and the negative correlation scheduling factors on the computing resource side, including the segmentation method, scheduling priority resource configuration, and corresponding execution effects.

[0060] Extract the correlation factors and historical performance indicators of segmentation granularity, scheduling order, and resource allocation strategy, and generate the optimal strategy through quantitative scoring. The comprehensive scoring formula is: Comprehensive Score = Keyframe Extraction Accuracy × 0.6 + 1 - Execution Time / Baseline Time × 0.4. Select the granularity scheme with the highest score. For example, for real-time video with frequent content switching, a 5-second small-granularity segment has an accuracy of 98% and a execution time of 12 seconds, so its score is higher than that of the 10-second medium-granularity segment. Finally, the small-granularity segmentation strategy is output.

[0061] The priority weight formula is: Priority Weight = Task Type Weight × 0.5 + Resource Adaptability × 0.3 + Historical Completion Rate × 0.2. Real-time live streaming segments are set as high priority for execution. Based on the resource adaptability formula: Resource Adaptability = Resource Utilization × 0.4 + Task Execution Efficiency Improvement Rate × 0.6, the optimal resource allocation is matched to segments with different computing power requirements. For example, 8 CPU cores and 16GB of memory are allocated to high computing power segments, and 4 CPU cores and 8GB of memory are allocated to low computing power segments to avoid resource waste and overload.

[0062] The optimal scheduling result for the intelligent extraction of keyframes to be processed is formed. The optimal scheduling result adapts to both video service requirements and real-time resource status. The scheduling effect prediction formula is: scheduling effect prediction value = keyframe extraction accuracy × 0.3 + task execution efficiency × 0.3 + resource utilization rate × 0.2 + task completion rate × 0.2, which ensures efficient and stable execution of the task.

[0063] The scheduling graph is constructed based on historical fragmented scheduling schemes and historical execution results, specifically including the following steps:

[0064] By decomposing historical sharding scheduling schemes, relevant sharding scheduling information is obtained; and by filtering historical execution results, effective execution result information is obtained.

[0065] Establish a basic association dataset by establishing a correspondence between information related to segmented scheduling and information on effective execution results;

[0066] Based on the type characteristics of keyframe extraction tasks, a keyframe extraction scheduling graph is constructed from the basic associated dataset, specifically including the following steps:

[0067] The basic association dataset is classified and organized according to the type characteristics of the keyframe extraction task to obtain the classification association data;

[0068] A keyframe extraction scheduling graph is constructed by combining specific scheduling information and execution effect information of the classified and associated data.

[0069] Information related to fragmented scheduling includes fragmentation methods, scheduling priorities, and resource allocation.

[0070] By decomposing the historical fragmented scheduling scheme, fragmented scheduling-related information is obtained, including fragmentation method, scheduling priority, and resource configuration. Among them, the segmentation method refers to the specific rules for segmenting video data in historical tasks. For example, a single video segment is divided into 10-second continuous segments according to the time dimension, or the video is divided into segments with drastic content changes and segments with gradual content changes according to the content complexity. The scheduling priority refers to the execution order of different segment tasks in historical scheduling. For example, the segment priority of the keyframe extraction task for live streaming is set higher than that of the offline storage video task. The resource configuration refers to the computing resources allocated to each segment task in historical scheduling. For example, 8 CPU cores and 16GB of memory are allocated to high-priority segments, and 4 CPU cores and 8GB of memory are allocated to low-priority segments. The historical execution effect is filtered to remove invalid data with execution anomalies, missing data, or that does not conform to the characteristics of the task type. Only valid execution effect information that can truly reflect the execution effect of the scheduling scheme is retained. For example, execution data that failed due to network interruption is removed, and only valid data that shows that the task was completed normally and the time and resource utilization met expectations are retained.

[0071] Establish a correspondence between information related to shard scheduling and information on effective execution results to form a basic association dataset. Bind each set of historical shard scheduling schemes to its corresponding effective execution results, clarifying the actual performance of different scheduling strategies. For example, associate the scheduling scheme of "10-second time sharding, high priority, 8 cores and 16GB resource configuration" with the corresponding execution results of "task execution time of 12 seconds, resource utilization of 85%, and keyframe extraction accuracy of 98%" to form a complete association data record. All such association data together constitute the basic association dataset.

[0072] A keyframe extraction scheduling graph is constructed based on the type characteristics of keyframe extraction tasks from the basic associated dataset. This involves classifying and organizing the basic associated dataset according to the type characteristics of keyframe extraction tasks, resulting in categorized associated data. The type characteristics of keyframe extraction tasks are divided based on dimensions such as application scenarios and video source attributes. For example, they are divided into real-time live streaming keyframe extraction tasks, offline video content keyframe extraction tasks, and security monitoring keyframe extraction tasks. Based on the scheduling requirements and execution characteristics of different task types, the associated data of corresponding types in the basic associated dataset are categorized and organized to form corresponding categorized associated data. For example, the scheduling and execution data of all real-time live streaming tasks are grouped into one category, and the scheduling and execution data of all offline video content tasks are grouped into another category.

[0073] A keyframe extraction scheduling graph is constructed by combining specific scheduling information and execution effect information from the categorized and associated data. The graph structure is built using scheduling factors as core nodes and the correlation trajectories between scheduling factors as edges. Each scheduling factor includes factor feature items and factor feature values. For example, the feature item could be the sharding method, the feature value could be a 10-second time shard, the feature item could be scheduling priority, the feature value could be high priority, the feature item could be resource configuration, the feature value could be 8 cores and 16GB of RAM, and the feature item could be execution time, with 12 seconds as the feature value. Correspondence trajectories between scheduling factors are established based on the correspondences in the associated data. For example, the correlation trajectory could be established between "10-second time sharding - high priority - 8 cores and 16GB of RAM - 12-second execution time," clarifying the influence relationships and action paths between different scheduling factors. Finally, a complete keyframe extraction scheduling graph is formed, presenting the characteristic performance, correlation relationships, and corresponding execution effects of various scheduling factors under different types of tasks.

[0074] In the process of data association and graph construction, the correlation strength between scheduling factors and execution results is quantified using a correlation degree calculation formula: Correlation Degree = Covariance of Scheduling Factor Feature Values ​​and Execution Results Feature Values ​​ / (Standard Deviation of Scheduling Factor Feature Values ​​× Standard Deviation of Execution Results Feature Values). This calculates the degree of influence of different scheduling factors on execution results. For example, the correlation degree between sharding method and execution time is 0.82, scheduling priority is 0.91, and resource allocation is 0.87. The weight of each scheduling factor is clearly defined in graph construction to enhance the graph's decision support capability. The classification accuracy is verified using a classification accuracy calculation formula: Classification Accuracy = Number of Correctly Classified Related Data Items / Total Number of Items in the Basic Related Data Set × 100%. This ensures the accuracy of the classified related data and guarantees the reliability of subsequent graph construction.

[0075] The process of processing real-time task feature data to generate real-time scheduling factors includes the following steps:

[0076] Real-time task feature data includes video source feature data and computing resource feature data;

[0077] The video scheduling factors are obtained based on the video source feature data, specifically including the following steps:

[0078] Video source regularization feature information is obtained from video source feature data; among which, video source regularization feature information includes effective information on video transmission status and the degree of content change;

[0079] The video state determination result is obtained by performing state determination on the regularized feature information of the video source; the video state determination result includes the stability of the video transmission process and the frequency of content switching.

[0080] Video scheduling factors that guide task scheduling execution are derived from video status determination results;

[0081] The resource scheduling factors are derived from computational resource characteristic data, specifically including the following steps:

[0082] The resource operation status perception results are obtained by performing operation status perception based on computing resource characteristic information; the resource operation status perception results include the load status and availability of computing resources;

[0083] Based on the results of resource operation situation awareness, the support capabilities and constraints of different resources for task execution are determined, and the resource adaptation analysis results are obtained.

[0084] Real-time task feature data includes video source feature data and computing resource feature data. The video source feature data reflects the video's inherent attributes and transmission status for the keyframe extraction task, while the computing resource feature data reflects the real-time operating status of the hardware and system resources executing the task.

[0085] Video scheduling factors are generated based on video source feature data. Video source regularization feature information is obtained based on video source feature data. The original video source data is cleaned, filtered and standardized to extract effective information on video transmission status and content change degree, and invalid noise data is removed. For example, transmission status information such as video bitrate, fluctuation range and packet loss rate, as well as content change degree information such as pixel difference between adjacent frames, change rate and scene switching number are extracted from the original video data to form structured video source regularization feature information. The video status is determined by regularizing the feature information of the video source. Based on the regularized feature information, the video's operating status is quantitatively evaluated. Specifically, this includes identifying the stability of the video transmission process and the frequency of content switching. For example, the transmission stability is determined by calculating the video bitrate fluctuation coefficient, where fluctuation coefficient = maximum bitrate - minimum bitrate / average bitrate. When the fluctuation coefficient is less than 0.1, the transmission is considered stable, and when the fluctuation coefficient is greater than 0.3, the transmission is considered unstable. The content switching frequency is determined by counting the number of scene switching times per unit time. For example, if the number of scene switching times per minute is less than 2, the content is considered smooth, and if the number of scene switching times per minute is greater than 5, the content is considered frequent. Finally, a video status determination result containing both the transmission stability level and the content switching level is formed.

[0086] Based on the video state determination results, video scheduling factors are obtained to guide task scheduling execution. The state determination results are transformed into guiding factors that can be used for scheduling decisions. For example, for videos with stable transmission and smooth content, video scheduling factors with low-priority segmentation and large-granularity segmentation are generated; for live videos with unstable transmission and frequent content switching, video scheduling factors with high-priority segmentation and small-granularity segmentation are generated. This ensures that the scheduling scheme adapts to the actual running state of the video and guarantees the real-time performance and accuracy of keyframe extraction.

[0087] Resource scheduling factors are generated based on computing resource characteristic data. Operational status perception results are obtained based on computing resource characteristic information. Real-time operation data of computing resources are comprehensively collected and analyzed to obtain the load and availability of computing resources. Load includes CPU utilization, memory utilization, disk I / O utilization, and network bandwidth utilization. Availability includes the number of remaining CPU cores, remaining memory capacity, remaining disk space, and remaining network bandwidth. For example, if a computing node has a CPU utilization of 75%, a memory utilization of 60%, 2 remaining CPU cores, and 8GB of remaining memory, the resource operation status perception results for that node are formed.

[0088] Based on the resource operation status awareness results, the support capabilities and constraints of different resources for task execution are determined, resulting in resource adaptation analysis results. The operational status of resources is evaluated to clarify the types and scale of tasks that each resource can support, as well as the constraints of resource operation. For example, for computing nodes with CPU utilization exceeding 80%, it is determined that their ability to support high-computing-power-demand tasks is insufficient, and the constraint is that they can only be allocated low-computing-power-demand sharding tasks. For nodes with remaining network bandwidth of less than 100Mbps, it is determined that their ability to support high-bitrate video tasks is insufficient, and the constraint is that they can only be allocated low-bitrate video sharding tasks. Finally, a resource adaptation analysis result containing resource support levels and constraints is formed.

[0089] Based on the resource adaptation analysis results, the resource scheduling factors that influence scheduling decisions are identified;

[0090] Based on the resource adaptation analysis results, the resource scheduling factors that influence scheduling decisions are identified. These results are then transformed into resource-side factors that can be used for scheduling decisions. For example, for computing nodes with strong support capabilities and no obvious constraints, resource scheduling factors with high resource allocation priority and large resource quotas are generated; for computing nodes with weak support capabilities and many constraints, resource scheduling factors with low resource allocation priority and small resource quotas are generated. This ensures that the scheduling scheme adapts to the real-time operating status of computing resources and avoids resource overload leading to task execution failures.

[0091] Combine video scheduling factors and resource scheduling factors into real-time scheduling factors;

[0092] The feature analysis results of real-time scheduling factors are obtained through feature analysis, specifically including the following steps:

[0093] Regularize the real-time scheduling factors to obtain regularized scheduling factors;

[0094] The correlation determination results of factors are obtained by performing correlation determination on the regularized scheduling factors;

[0095] The factor hierarchy division results based on the factor association discrimination results are used to determine the degree of execution effect;

[0096] Based on the factor hierarchy classification results, various scheduling factors are refined to obtain feature analysis results for scheduling matching. Based on the feature analysis results and the scheduling factor library, real-time scheduling factors corresponding to real-time scheduling factors are generated.

[0097] Video scheduling factors and resource scheduling factors are combined to form real-time scheduling factors. The two types of scheduling factors generated from the video service side and the computing resource side are integrated to form a comprehensive set of scheduling factors that include all dimensions of task characteristics. Among them, video scheduling factors carry the scheduling requirements related to video attributes and transmission status of the keyframe extraction task to be processed, while resource scheduling factors carry the scheduling constraints related to the real-time operating status and adaptability of computing resources.

[0098] Feature analysis is performed on real-time scheduling factors to obtain feature analysis results. These factors are then regularized to obtain regularized scheduling factors. The integrated real-time scheduling factors undergo format unification, dimension alignment, and noise removal to eliminate format differences and redundant information from different sources, transforming various scheduling factors into standardized data with unified dimensions and formats. For example, numerical indicators such as transmission stability and content switching intensity in video scheduling factors are aligned with numerical indicators such as comprehensive load rate and resource availability in resource scheduling factors, uniformly represented by standardized values ​​from 0 to 1. Invalid and abnormal data are also removed, resulting in regularized scheduling factors with consistent structure and unified dimensions. The regularization process uses a standardized calculation formula: Standardized value = Original value - Minimum value / Maximum value - Minimum value.

[0099] The correlation results are obtained by performing correlation analysis on the normalized scheduling factors. Based on the normalized standardized data, the correlation degree between different scheduling factors is quantitatively calculated to clarify the mutual influence relationships between factors. The correlation degree can be calculated using the Pearson correlation coefficient formula: Correlation degree = Covariance of factor A and factor B / Standard deviation of factor A × Standard deviation of factor B. This formula calculates the correlation strength between each pair of scheduling factors; the closer the value is to 1, the stronger the positive correlation; the closer it is to -1, the stronger the negative correlation; and the closer it is to 0, the less significant the correlation. For example, the correlation degree between video content switching intensity and segment granularity requirement is calculated to be 0.89, indicating that the more drastic the content switching, the stronger the demand for small-granularity segments; the correlation degree between overall resource load rate and resource allocation priority is calculated to be -0.92, indicating that the higher the resource load, the lower the allocation priority. Finally, a factor correlation analysis result encompassing the correlation strength between all scheduling factors is formed.

[0100] The hierarchical classification of factors affecting scheduling execution is based on the correlation discrimination results. According to the correlation discrimination results, all scheduling factors are hierarchically classified according to their degree of influence and level of effect on scheduling execution, clarifying core influencing factors, secondary influencing factors, and auxiliary influencing factors. For example, strongly correlated factors with a correlation degree higher than 0.7 are classified as core influencing factors, factors with a correlation degree between 0.4 and 0.7 are classified as secondary influencing factors, and factors with a correlation degree lower than 0.4 are classified as auxiliary influencing factors. Furthermore, according to the level of influence of business and resources, core factors on the video side are classified as business level, and core factors on the resource side are classified as resource level, thus forming a hierarchical structure of factors and clarifying the weight and order of influence of each factor in scheduling decisions. For example, video content switching intensity and transmission stability are classified as core factors at the business level, resource comprehensive load rate and resource availability are classified as core factors at the resource level, and weakly correlated factors such as disk I / O utilization are classified as auxiliary influencing factors, ultimately obtaining a complete hierarchical classification of factors.

[0101] Based on the hierarchical classification of factors, feature analysis results for scheduling matching are extracted from various scheduling factors. The process involves refining scheduling factors based on the hierarchical classification, retaining key influencing factors at the core level, and eliminating redundant auxiliary factors to form feature analysis results. This ensures the accuracy and efficiency of subsequent scheduling matching. For example, core influencing factors at the business and resource levels are retained, while weakly correlated factors at the auxiliary level are eliminated. The standardized values ​​of the core factors are used as the core content of the feature analysis results, and the hierarchical weight of each core factor is clearly defined.

[0102] Based on the feature analysis results and the scheduling factor library, real-time scheduling factors corresponding to real-time scheduling factors are generated. The feature analysis results are matched and mapped with a pre-defined scheduling factor library, which stores standardized scheduling factors. Each scheduling factor includes scheduling feature items and scheduling feature values; for example, sharding granularity can be used as the scheduling feature item, smaller granularity as the scheduling feature value, resource allocation priority as the scheduling feature item, and high priority as the scheduling feature value. By mapping the core factors of the feature analysis results to the feature items in the scheduling factor library, the feature values ​​are mapped to the corresponding feature values, ultimately generating real-time scheduling factors that perfectly match the real-time scheduling factors.

[0103] The target scheduling factor is obtained by matching the real-time scheduling factor with the scheduling graph, specifically including the following steps:

[0104] The scheduling feature terms of the real-time scheduling factor are matched with the factor feature terms of the scheduling factor in the scheduling graph to obtain the matching scheduling factor corresponding to the real-time scheduling factor.

[0105] The target scheduling factor corresponding to the real-time scheduling factor is obtained by matching the scheduling feature value of the real-time scheduling factor with the factor feature value of the matching scheduling factor.

[0106] The scheduling features of real-time scheduling factors are matched with the factor features of scheduling factors in the scheduling graph to obtain matching scheduling factors corresponding to the real-time scheduling factors. Scheduling features and factor features are classification identifiers for scheduling factors, used to define the type of scheduling attributes carried by the scheduling factors. For example, segmentation granularity scheduling priority, resource allocation quota, video transmission stability, and overall resource load rate all fall under the category of features. A set of candidate scheduling factors matching the real-time scheduling factor features is selected from the scheduling graph to form matching scheduling factors. For example, if the scheduling feature of a real-time scheduling factor is segmentation granularity, the feature matching process traverses all scheduling factors in the scheduling graph, retaining only those with segmentation granularity as matching scheduling factors for that real-time scheduling factor, eliminating irrelevant scheduling factors with mismatched features, significantly narrowing the subsequent matching range and improving matching efficiency. Feature matching adopts a complete matching rule, meaning that a feature match is considered successful only when the scheduling features of a real-time scheduling factor are completely identical to the factor features of scheduling factors in the scheduling graph.

[0107] The scheduling feature values ​​of the real-time scheduling factor are matched with the factor feature values ​​of the matching scheduling factor to obtain the target scheduling factor corresponding to the real-time scheduling factor. The scheduling feature values ​​and factor feature values ​​are specific quantitative or qualitative values ​​of corresponding feature items. For example, the feature values ​​corresponding to the sharding granularity feature item are divided into small granularity, medium granularity, and large granularity, or corresponding to specific sharding durations such as 5 seconds, 10 seconds, and 20 seconds; the feature values ​​corresponding to the resource comprehensive load rate feature item are divided into low load, medium load, and high load, or corresponding to specific load values ​​such as 30%, 60%, and 90%. Based on feature item alignment, the matching of specific attribute values ​​is completed. The scheduling factor that completely matches or has the highest similarity to the real-time scheduling factor feature value is selected from the matched scheduling factors as the final target scheduling factor. Feature value matching uses a similarity calculation formula to quantify the degree of matching: Similarity = 1 - Feature value difference / Feature value baseline range. Here, the feature value difference is the absolute difference between the real-time scheduling factor feature value and the matching scheduling factor feature value, and the feature value baseline range is the difference between the maximum and minimum values ​​of the feature item. The closer the similarity value is to 1, the higher the degree of matching. For example, if the granularity feature value of the real-time scheduling factor is 5 seconds, and there are scheduling factors with granularity feature values ​​of 5 seconds and 10 seconds among the matching scheduling factors, the similarity of the 5-second feature value is calculated to be 1, and the similarity of the 10-second feature value is 0.5. Therefore, the scheduling factor with a feature value of 5 seconds is selected as the target scheduling factor.

[0108] Determining the corresponding target-associated scheduling factor set based on the target scheduling factor specifically includes the following steps:

[0109] If the real-time scheduling factor corresponding to the target scheduling factor is a video scheduling factor, then obtain the positive pointing mark of the target scheduling factor, obtain the positive correlation trajectory of the target scheduling factor based on the positive pointing mark, and determine all scheduling factors of the positive correlation trajectory as the target positive scheduling factor.

[0110] If the real-time scheduling factor corresponding to the target scheduling factor is a resource scheduling factor, then obtain the reverse pointing mark of the target scheduling factor, obtain the reverse correlation trajectory of the target scheduling factor based on the reverse pointing mark, and determine all scheduling factors in the reverse correlation trajectory as the target reverse scheduling factor.

[0111] The target forward scheduling factor and / or the target reverse scheduling factor constitute the target associated scheduling factor set.

[0112] If the real-time scheduling factor corresponding to the target scheduling factor is a video scheduling factor, then the forward correlation trajectory tracing process is executed. The video scheduling factor carries the video service-side requirements of the keyframe extraction task to be processed. It corresponds to a forward correlation link in the scheduling graph starting from the service requirement. The forward pointing marker is a pre-defined correlation direction identifier in the scheduling graph for video scheduling factors, used to indicate the correlation path from the service requirement factor to the corresponding scheduling strategy and execution effect factor. The forward pointing marker of the target scheduling factor is obtained, and then the scheduling graph is traversed according to the forward pointing marker to extract the forward correlation trajectory of the target scheduling factor. The forward correlation trajectory is a link consisting of all scheduling factors extending along the forward pointing marker from the target scheduling factor, including historical segmentation scheduling schemes and execution effect information related to this video scheduling factor. For example, the target scheduling factor is the scheduling factor corresponding to frequent video content switching. Its positive pointing marker indicates that it starts from the business demand factor, is associated with the scheduling strategy factor with high scheduling priority and high resource allocation in small-granular segmentation, and is then associated with the execution effect factor with corresponding execution time and key frame extraction accuracy. All scheduling factors in the positive association trajectory are determined as the target positive scheduling factor, forming a complete set of business-side associated scheduling factors.

[0113] If the real-time scheduling factor corresponding to the target scheduling factor is a resource scheduling factor, then the reverse correlation trajectory tracing process is executed. Resource scheduling factors carry the real-time operating status and constraints of the computing resources side, corresponding to a reverse correlation link in the scheduling graph starting from the resource status. The reverse pointing marker is a preset correlation direction identifier for resource-type scheduling factors in the scheduling graph, used to indicate the correlation path from the resource status factor to the corresponding adapted scheduling strategy and execution effect factor. First, the reverse pointing marker of the target scheduling factor is obtained. Then, the scheduling graph is traversed according to the reverse pointing marker to extract the reverse correlation trajectory of the target scheduling factor. The reverse correlation trajectory is a link consisting of all scheduling factors extending along the reverse pointing marker from the target scheduling factor, including historical sharding scheduling schemes and execution effect information related to the resource scheduling factor. For example, if the target scheduling factor is a scheduling factor corresponding to high resource load, its reverse pointing marker indicates that it starts from the resource status factor, is associated with a scheduling strategy factor with low scheduling priority and small resource quota, and then associated with the execution effect factor with corresponding execution time and resource utilization. All scheduling factors in the reverse correlation trajectory are determined as the target reverse scheduling factor, forming a complete set of resource-side associated scheduling factors.

[0114] The target-related scheduling factor set is constructed by combining the target forward scheduling factor and / or the target reverse scheduling factor. If only the target scheduling factor corresponding to the video scheduling factor is matched, the set consists only of the target forward scheduling factor; if only the target scheduling factor corresponding to the resource scheduling factor is matched, the set consists only of the target reverse scheduling factor; if the target scheduling factors corresponding to both the video scheduling factor and the resource scheduling factor are matched, the set consists of both the target forward scheduling factor and the target reverse scheduling factor, achieving a comprehensive integration of scheduling knowledge between the business side and the resource side. The target-related scheduling factor set includes all historical scheduling factors related to the currently pending task, containing the corresponding scheduling strategies and execution effect information.

[0115] During the trajectory tracing process, the correlation strength calculation formula quantifies the degree of correlation among scheduling factors within the trajectory: Correlation Strength = Historical Correlation Count between Scheduling Factors / Total Correlation Count. A higher correlation strength value indicates a closer correlation between the scheduling factor and the target scheduling factor in historical scheduling, thus assigning it a higher weight in the target correlated scheduling factor set and improving the accuracy of subsequent scheduling decisions. The effectiveness of the correlated trajectories is verified through the trajectory integrity calculation formula: Trajectory Integrity = Number of Effective Scheduling Factors within the Trajectory / Total Number of Scheduling Factors in the Trajectory × 100%. This ensures the integrity and reliability of the target correlated scheduling factor set, providing a solid guarantee for the generation of optimal scheduling results.

[0116] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0117] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0118] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A keyframe intelligent extraction task optimization system based on dynamic fragmentation scheduling, characterized in that, include: Acquisition module: Acquires historical segmentation scheduling schemes and historical execution results of keyframe extraction tasks; acquires real-time task feature data of keyframe extraction tasks to be processed; Extraction module: Constructs a keyframe extraction scheduling graph based on historical fragmentation scheduling schemes and historical execution effects. The scheduling graph includes scheduling factors and the associated trajectories between scheduling factors. The scheduling factors include factor feature terms and factor feature values. Generation module: processes real-time task feature data to generate real-time scheduling factors; wherein, the real-time scheduling factors include scheduling feature items and scheduling feature values; Matching module: Matches real-time scheduling factors with the scheduling graph to obtain target scheduling factors; determines the corresponding set of target associated scheduling factors based on the target scheduling factors; Output module: Based on the target associated scheduling factor set, output the fragmentation granularity, scheduling order and resource allocation strategy to form the optimal scheduling result of the keyframe intelligent extraction task to be processed.

2. The keyframe intelligent extraction task optimization system based on dynamic fragmentation scheduling according to claim 1, characterized in that, The scheduling graph is constructed based on historical fragmented scheduling schemes and historical execution results, specifically including the following steps: By decomposing historical sharding scheduling schemes, relevant sharding scheduling information is obtained; and by filtering historical execution results, effective execution result information is obtained. Establish a basic association dataset by establishing a correspondence between information related to segmented scheduling and information on effective execution results; A keyframe extraction scheduling graph is constructed based on the type characteristics of the keyframe extraction task and the basic associated dataset.

3. The keyframe intelligent extraction task optimization system based on dynamic fragmentation scheduling according to claim 2, characterized in that, The information related to the fragmentation scheduling includes the fragmentation method, scheduling priority, and resource configuration.

4. The keyframe intelligent extraction task optimization system based on dynamic fragmentation scheduling according to claim 1, characterized in that, The process of processing real-time task feature data to generate real-time scheduling factors includes the following steps: The real-time task feature data includes video source feature data and computing resource feature data; Video scheduling factors are obtained based on video source feature data, and resource scheduling factors are obtained based on computing resource feature data. Combine video scheduling factors and resource scheduling factors into real-time scheduling factors; Feature analysis is performed on real-time scheduling factors to obtain feature analysis results. Based on the feature analysis results and the scheduling factor library, real-time scheduling factors corresponding to the real-time scheduling factors are generated.

5. The keyframe intelligent extraction task optimization system based on dynamic fragmentation scheduling according to claim 4, characterized in that, The video scheduling factors are obtained based on the video source feature data, specifically including the following steps: Video source regularization feature information is obtained based on video source feature data; wherein, the video source regularization feature information includes effective information on video transmission status and the degree of content change; The video state determination result is obtained by performing state determination on the regularized feature information of the video source; wherein, the video state determination result includes the stability of the video transmission process and the frequency of content switching. Video scheduling factors that guide task scheduling execution are derived from video status determination results.

6. The keyframe intelligent extraction task optimization system based on dynamic fragmentation scheduling according to claim 4, characterized in that, The resource scheduling factors are derived from computational resource characteristic data, specifically including the following steps: The resource operation status perception results are obtained by performing operation status perception based on computing resource characteristic information; the resource operation status perception results include the load and availability of computing resources; Based on the results of resource operation situation awareness, the support capabilities and constraints of different resources for task execution are determined, and the resource adaptation analysis results are obtained. Based on the resource adaptation analysis results, the resource scheduling factors that influence scheduling decisions are identified.

7. The keyframe intelligent extraction task optimization system based on dynamic fragmentation scheduling according to claim 6, characterized in that, The feature analysis results of real-time scheduling factors are obtained through feature analysis, specifically including the following steps: Regularized scheduling factors are obtained by regularizing real-time scheduling factors. The correlation determination results of factors are obtained by performing correlation determination on the regularized scheduling factors; The factor hierarchy division results based on the factor association discrimination results are used to determine the degree of execution effect; Based on the result of factor hierarchy classification, various scheduling factors are extracted to obtain feature analysis results for scheduling matching.

8. The keyframe intelligent extraction task optimization system based on dynamic fragmentation scheduling according to claim 1, characterized in that, The target scheduling factor is obtained by matching the real-time scheduling factor with the scheduling graph, specifically including the following steps: The scheduling feature terms of the real-time scheduling factor are matched with the factor feature terms of the scheduling factor in the scheduling graph to obtain the matching scheduling factor corresponding to the real-time scheduling factor. The target scheduling factor corresponding to the real-time scheduling factor is obtained by matching the scheduling feature value of the real-time scheduling factor with the factor feature value of the matching scheduling factor.

9. The keyframe intelligent extraction task optimization system based on dynamic fragmentation scheduling according to claim 1, characterized in that, Determining the corresponding target-associated scheduling factor set based on the target scheduling factor specifically includes the following steps: If the real-time scheduling factor corresponding to the target scheduling factor is a video scheduling factor, then obtain the positive pointing mark of the target scheduling factor, obtain the positive correlation trajectory of the target scheduling factor based on the positive pointing mark, and determine all scheduling factors of the positive correlation trajectory as the target positive scheduling factor. If the real-time scheduling factor corresponding to the target scheduling factor is a resource scheduling factor, then obtain the reverse pointing mark of the target scheduling factor, obtain the reverse correlation trajectory of the target scheduling factor based on the reverse pointing mark, and determine all scheduling factors in the reverse correlation trajectory as the target reverse scheduling factor. The target forward scheduling factor and / or the target reverse scheduling factor constitute the target associated scheduling factor set.

10. The keyframe intelligent extraction task optimization system based on dynamic fragmentation scheduling according to claim 2, characterized in that, Based on the type characteristics of keyframe extraction tasks, a keyframe extraction scheduling graph is constructed from the basic associated dataset, specifically including the following steps: The basic association dataset is classified and organized according to the type characteristics of the keyframe extraction task to obtain the classification association data; A keyframe extraction scheduling graph is constructed by combining specific scheduling information and execution effect information of the classified and associated data.

Images