High-speed train image snapshot processing system and method based on AI

Abstract

The invention relates to the technical field of image processing, and discloses an AI-based high-speed train image snapshot processing system and method, and the method comprises the steps: carrying out the processing of continuously obtained high-speed train images based on a sliding window algorithm; a dynamic association graph network model is constructed, and a time sequence graph attention network is applied to carry out target space relation processing; using a multi-hop anomaly correlation analysis algorithm to predict propagation of anomaly between targets; a self-adaptive resource allocation system is realized, and image acquisition parameters are dynamically adjusted according to a scene importance evaluation model; executing an intelligent cache optimization algorithm to realize value-based data management and storage resource optimization distribution; and through dynamic association graph network modeling and multi-hop anomaly association analysis, the anomaly early warning accuracy is improved.

Description

Technical Field

[0001] The present invention relates to the field of image processing technology, and more specifically, to an AI-based high-speed train image capture and processing system and method. Background Art

[0002] The high-speed train monitoring system generates massive amounts of image data every day. These images contain information about equipment, track infrastructure, and environmental conditions along the line. Traditional image capture methods use fixed-interval sampling, which cannot intelligently identify key scenes, resulting in wasted storage space and missing important information. At the same time, the operating environment of high-speed trains is complex, and there are potential correlations between various monitoring targets. When an abnormality occurs in one device, it may trigger a chain reaction that affects other related devices. This correlation cannot be effectively captured in traditional independent image processing methods.

[0003] The current image monitoring technology has the following main problems: a large number of redundant images occupying storage space; key scene missed detection; unreasonable resource allocation; event correlation identification; key diagnostic information omission. These problems lead to low efficiency of the monitoring system and cannot effectively support the safe operation and maintenance of high-speed trains. Summary of the invention

[0004] The present invention provides an AI-based high-speed train image capture and processing system and method, which solves the technical problems of redundant data storage and omission of key information in traditional image monitoring technology in related technologies.

[0005] The present invention provides an AI-based high-speed train image capture processing method, comprising:

[0006] The continuously acquired high-speed train images are processed based on a sliding window algorithm to identify key changes in the scene;

[0007] Using the key changes in the scene, a dynamic association graph network model is constructed and a temporal graph attention network is applied to process the target spatial relationship and generate a target association graph;

[0008] Based on the target association graph, a multi-hop anomaly association analysis algorithm is used to predict the propagation of anomalies between targets and obtain the probability matrix of anomaly propagation paths;

[0009] Based on the key changes in the scene and the probability matrix of abnormal propagation paths, an adaptive resource allocation system is implemented to dynamically adjust image acquisition parameters according to the scene importance evaluation model;

[0010] According to the results of the anomaly propagation path probability matrix and the adaptive resource allocation system, an intelligent cache optimization algorithm is executed to achieve value-based data management and storage resource optimization allocation.

[0011] Furthermore, the processing of continuously acquired high-speed train images based on a sliding window algorithm includes:

[0012] The sliding window algorithm is used to maintain the latest N frame image sequence;

[0013] Calculate the information entropy difference between adjacent image frames;

[0014] Calculate the structural similarity between adjacent image frames;

[0015] The scene change index is constructed by combining information entropy difference and structural similarity.

[0016] Furthermore, the construction of the dynamic association graph network model includes:

[0017] Use object detection algorithms to identify key sets of objects from images;

[0018] Construct scene dynamic association graph;

[0019] Calculate the association weights between targets;

[0020] Applying temporal graph attention network to process dynamic correlations between objects.

[0021] Furthermore, the method of using a multi-hop anomaly association analysis algorithm to predict the propagation of anomalies between targets includes:

[0022] Calculate the adjacency matrix and degree matrix corresponding to the target association graph;

[0023] Construct a normalized adjacency matrix;

[0024] Initialize the node’s abnormal attention vector;

[0025] The attention propagation algorithm is used to iteratively calculate the k-hop anomaly associations;

[0026] Construct an anomaly propagation model to predict the diffusion path of anomalies on the target graph.

[0027] Furthermore, the adaptive resource allocation system includes:

[0028] Based on the scene change indicators, a scene importance assessment model is constructed;

[0029] Dynamically adjust the temporal resolution and adaptively allocate sampling frequency according to the importance of the scene;

[0030] Build a hierarchical cache structure;

[0031] When an important event is detected, the retrospective analysis algorithm is triggered to temporarily increase the density of historical frame retention.

[0032] Furthermore, the intelligent cache optimization algorithm includes:

[0033] Optimize storage resource allocation strategy;

[0034] Build an intelligent elimination algorithm based on information value to evaluate the retention value of each frame of image;

[0035] Combined with storage resource constraints, an optimized elimination algorithm is implemented;

[0036] Implementing incremental learning models by continuously monitoring system performance indicators;

[0037] Build a cache access index system.

[0038] Furthermore, the calculation formula for constructing the scene change index is:

[0039] ;

[0040] in Indicates time The scene change indicator, and They represent the weight coefficients of information entropy difference and structural similarity in the scene change index, represents the information entropy difference between adjacent image frames, represents the structural similarity index, , Respectively indicate time ,time of the image.

[0041] Furthermore, the calculation formula for the dynamic association between the targets processed by the time sequence graph attention network is:

[0042] ;

[0043] in is the scene dynamic association diagram, is the hidden state at the current moment, containing the feature representations of all nodes; is the observed feature at the current moment, that is, the target feature extracted from the image; represents the temporal graph attention network, which combines graph neural networks and attention algorithms to capture the correlation information between objects and maintain temporal coherence; is the updated hidden state, which contains temporal and spatial relationship information.

[0044] Furthermore, the calculation formula for constructing the anomaly propagation model and predicting the diffusion path of the anomaly on the target graph is:

[0045] ;

[0046] in Indicates that at the node Under abnormal conditions, the node The conditional probability of anomaly. Represents a slave node To Node The abnormal propagation attention value of and Respectively represent nodes and nodes Status characteristics; Calculates the conditional probability function.

[0047] The present invention provides an AI-based high-speed train image capture processing system, which is used to execute the above-mentioned AI-based high-speed train image capture processing method, including:

[0048] Multi-dimensional information entropy analysis module, used to process high-speed train images based on a sliding window algorithm and identify key changes in scenes;

[0049] The dynamic association graph construction module is used to construct the spatial relationship network between targets based on the key change frames and generate the target association graph containing time series information;

[0050] The anomaly propagation prediction module is used to calculate the anomaly diffusion path between targets through multi-hop anomaly correlation analysis to form a risk probability matrix;

[0051] Adaptive resource allocation module, which is used to dynamically adjust monitoring parameters according to the importance of the scene and the risk of abnormality, and improve the data quality of key areas;

[0052] The intelligent cache management module is used to optimize storage strategies based on abnormal propagation risks and information value assessment to achieve value-oriented management of data.

[0053] The beneficial effects of the present invention are: through multi-dimensional information entropy analysis, dynamic association graph network, multi-hop abnormal association analysis, adaptive resource allocation and intelligent cache optimization, efficient processing of high-speed train image capture is achieved;

[0054] The storage efficiency and key scenario capture rate have been improved, and the data storage demand has been reduced. The anomaly detection capability has been enhanced, and the accuracy of anomaly warning has been improved through dynamic association graph network modeling and multi-hop anomaly association analysis. The resource utilization efficiency has been optimized, the computing resource requirements have been reduced, and the system response speed has been improved. The diagnostic information integrity has been improved, and the accuracy of root cause analysis has been improved. The system scalability has been enhanced, and a modular design has been adopted to support the continuous optimization and expansion of system functions. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 It is a flow chart of a high-speed train image capture processing method based on AI of the present invention;

[0056] Figure 2 is a flow chart of step 1 in the present invention;

[0057] Figure 3 is a flow chart of step 2 in the present invention;

[0058] Figure 4 is a flow chart of step 3 in the present invention;

[0059] Figure 5 is a flow chart of step 4 in the present invention;

[0060] Figure 6 It is a flow chart of step 5 in the present invention. DETAILED DESCRIPTION

[0061] The subject matter described herein will now be discussed with reference to example implementations. It should be understood that the discussion of these implementations is only to enable those skilled in the art to better understand and implement the subject matter described herein, and the functions and arrangements of the elements discussed may be changed without departing from the scope of protection of the present specification. Various examples may omit, replace, or add various processes or components as needed. In addition, the features described in some examples may also be combined in other examples.

[0062] At least one embodiment of the present invention discloses a high-speed train image capture processing method based on AI, such as Figures 1 to 6 As shown, the following steps are included:

[0063] Step 1: Process the continuously acquired high-speed train images based on the sliding window algorithm to identify key changes in the scene;

[0064] This step constructs a sliding window to maintain the most recent N frames in the sequence, and calculates the information entropy difference and structural similarity between adjacent frames, thereby generating a scene change index for identifying key scenes. The specific implementation is as follows:

[0065] Step 1.1, using sliding window algorithm to maintain the latest N frame image sequence;

[0066] The most recent N-frame image sequence is represented as:

[0067] ;

[0068] in , , It represents the image at the current time t, the earliest image in the window, and the second earliest image in the window, and N is the window size;

[0069] Step 1.2, calculating the information entropy difference between adjacent image frames;

[0070] The calculation formula of information entropy difference is:

[0071] ;

[0072] in represents the information entropy difference between adjacent image frames, Indicates time images, and Respectively represent images and Information entropy of

[0073] The calculation formula is:

[0074] ;

[0075] in Indicates that the gray value in the image is The probability of a pixel appearing; represents the logarithm with base 2; For all gray values Sum from 0 to 255;

[0076] Step 1.3, calculating the structural similarity between adjacent image frames;

[0077] The similarity is expressed as:

[0078] ;

[0079] in represents the structural similarity index, and Respectively indicate time and time images, , Respectively represent images , The mean of , Respectively represent images , The standard deviation of Representing images and The covariance of and They represent the constant terms added to avoid the denominator being zero;

[0080] Step 1.4, combining information entropy difference and structural similarity to construct a scene change index;

[0081] The calculation formula of the scene change index is:

[0082] ;

[0083] in Indicates time The scene change indicator, represents the information entropy difference between adjacent image frames, represents the structural similarity index, and Respectively indicate time and time images, and They represent the weight coefficients of information entropy difference and structural similarity in the scene change index respectively;

[0084] The output of this step is the scene change index between each pair of adjacent image frames. The larger the index value, the greater the scene change, which can be used for subsequent key frame judgment. In addition, a series of key change frames are identified through threshold judgment:

[0085] ;

[0086] in , , Respectively represent the first, second, and Key change frames, is the number of key change frames;

[0087] These key change frames will serve as the core input of the dynamic association graph network model in step 2, and the scene change index It will be used as one of the basic inputs for building the scene importance assessment model in step 4. This intelligent screening mechanism based on information entropy is the basis for the entire system to achieve efficient storage and accurate analysis, and lays the data foundation for all subsequent steps.

[0088] Step 2: Using the key changes in the scene, a dynamic association graph network model is constructed and a temporal graph attention network is applied to process the target spatial relationship to generate a target association graph.

[0089] This step uses the key change frames identified in step 1 as input to build a dynamic association graph network model to represent the spatial relationship between objects and perform association analysis. The key change frames output in step 1 provide the key time segments that the system focuses on, greatly reducing the amount of data processed while ensuring that no important information is missed. The specific implementation is as follows:

[0090] Step 2.1, using the target detection algorithm to identify the key target set from the image;

[0091] The key target set is expressed as:

[0092] ;

[0093] in , , Respectively represent the first, second, and detected targets, Indicates the number of detected targets;

[0094] Step 2.2, construct a scene dynamic association graph;

[0095] The scene dynamic association graph is represented as:

[0096] ;

[0097] in Dynamic association diagram for the scene:

[0098] ;

[0099] in is a node set, , , Respectively represent the first, second, and target nodes, Indicates the number of detected targets;

[0100] ;

[0101] in is a set of edges, representing the potential relationship between targets; Representation Node With Node The edge between and is the node index;

[0102] ;

[0103] in is the weight set, Representation Node With Node the strength of the association between

[0104] Step 2.3, calculate the association weights between targets;

[0105] The weight calculation takes into account the following factors:

[0106] Spatial distance:

[0107] ;

[0108] in Representation Node and nodes The spatial distance between and Respectively represent nodes and nodes The location coordinates of Represents the Euclidean distance between two location coordinates;

[0109] Functional relevance:

[0110] Representation Node With Node the degree of functional connection;

[0111] History Interaction:

[0112] Representation Node With Node frequency of interaction in history;

[0113] The calculation formula of the association weight is:

[0114] ;

[0115] in , , represent the weight coefficients of spatial distance, functional relevance, and historical interaction, respectively; It is the influence range parameter of spatial distance, which controls the rate of distance attenuation; is an exponential function;

[0116] Step 2.4, apply the temporal graph attention network to process the dynamic associations between targets;

[0117] The calculation formula is:

[0118] ;

[0119] in is the scene dynamic association diagram, is the hidden state at the current moment, containing the feature representations of all nodes; is the observed feature at the current moment, that is, the target feature extracted from the image; represents the temporal graph attention network, which combines graph neural networks and attention algorithms to capture the correlation information between objects and maintain temporal coherence; is the updated hidden state, which contains temporal and spatial relationship information.

[0120] The output of this step is the target association graph and its updated node representation , these outputs are directly used as the basic data structure and computing resources for multi-hop abnormal association analysis in step 3. This association graph not only represents the spatial association relationship between targets, but also captures the dynamic changes of associations in the time dimension through the time series graph attention network, providing a complete relationship network model for subsequent abnormal association analysis. In addition, the node representation It contains rich target feature information, which will be further used in steps 3 and 4 to form a progressive relationship of the system's information flow, thereby ensuring the coherence of the entire processing flow and the consistency of information transmission.

[0121] Step 3: Based on the target association graph, use the multi-hop anomaly association analysis algorithm to predict the propagation of anomalies between targets and obtain the anomaly propagation path probability matrix;

[0122] This step makes full use of the target association graph output in step 2. and node representation On this basis, multi-hop anomaly association analysis is performed to predict the diffusion path of anomalies on the target graph. The association graph constructed in step 2 provides a complete target relationship network for anomaly propagation analysis, so that this step can accurately simulate the propagation process of anomalies. The specific implementation is as follows:

[0123] Step 3.1, calculate the adjacency matrix and degree matrix corresponding to the target association graph;

[0124] Based on the target association graph constructed in step 2 , calculate its corresponding adjacency matrix Sum degree matrix :

[0125] Adjacency Matrix:

[0126] ;

[0127] in is the adjacency matrix; express OK A real matrix of columns; is the field of real numbers;

[0128] ;

[0129] in Representation Node and nodes The edge weight between nodes and nodes If there is no connection between ;

[0130] Degree matrix:

[0131] ;

[0132] in is the degree matrix;

[0133] ;

[0134] in Representation Node degree, that is, the degree of the node The sum of the weights of all connected edges; Express From 1 to sum; Representation Node and nodes The edge weights between

[0135] Step 3.2, construct a normalized adjacency matrix;

[0136] The expression of the normalized adjacency matrix is:

[0137] ;

[0138] in is the normalized adjacency matrix, is the adjacency matrix; for The identity matrix has diagonal elements that are 1 and the rest of the elements are 0;

[0139] Construct the normalized degree matrix:

[0140] ;

[0141] in is the normalized degree matrix, is the degree matrix;

[0142] Step 3.3, initialize the node’s abnormal attention vector:

[0143] The initial abnormal attention vector is expressed as:

[0144] ;

[0145] in is the initial abnormal attention vector; express dimensional real vector space;

[0146] Step 3.4, use the attention propagation algorithm to iteratively calculate the k-hop anomaly association;

[0147] The calculation formula is:

[0148] ;

[0149] in is an activation function (such as ReLU), which is used to introduce nonlinear transformation; Represents the normalized degree matrix The negative power of one half, that is, the diagonal elements are raised to the negative power of one half; For the The weight matrix of the layer is used to learn the influence transfer pattern between different nodes; Indicates Abnormal attention vector after iterations; Indicates Abnormal attention vector after iterations;

[0150] Step 3.5, construct an anomaly propagation model to predict the diffusion path of anomalies on the target graph;

[0151] The calculation formula is:

[0152] ;

[0153] in Indicates that at the node Under abnormal conditions, the node The conditional probability of anomaly. Represents a slave node To Node The abnormal propagation attention value of and Respectively represent nodes and nodes Status characteristics; is the conditional probability calculation function, which can be in the following form:

[0154] ;

[0155] in is the state compatibility function, which is used to measure the node Status and nodes The compatibility of the state in the exception propagation; is an exponential function; Indicates that All except Perform the summation.

[0156] The output of this step is the anomaly propagation path probability matrix, where Indicates an abnormal slave node Propagate to nodes This probability matrix is ​​the core reference data for resource allocation and cache optimization in subsequent steps. It directly affects the scene importance assessment in step 4, because scenes predicted to be high-risk areas for abnormal propagation will be given higher importance weights; at the same time, it is also a key input for intelligent cache optimization in step 5, providing a risk-based priority basis for storage resource allocation. In this way, the system can prioritize the allocation of limited computing and storage resources to the areas most likely to have abnormalities, thereby achieving efficient capture of key abnormalities.

[0157] Step 4: Based on the key changes in the scene and the probability matrix of abnormal propagation paths, an adaptive resource allocation system is implemented to dynamically adjust the image acquisition parameters according to the scene importance evaluation model;

[0158] This step is based on the output of the previous steps, especially the scene change index of step 1. And the abnormal propagation path probability matrix of step 3 , build an adaptive resource allocation system, and dynamically optimize the time resolution and caching strategy of the monitoring system. This design that relies on the previous steps ensures that resource allocation decisions fully consider scene changes and the risk of abnormal propagation, making system resource utilization more efficient. The specific implementation is as follows:

[0159] Step 4.1, construct a scene importance assessment model based on scene change indicators;

[0160] The calculation formula of scene importance is:

[0161] ;

[0162] in Indicates time The importance of the scene, Indicates time The scene change indicator, , , The weight coefficients representing the scene change index, abnormality degree and context importance respectively; Indicates time The degree of abnormality can be extracted from the abnormal propagation path probability matrix in step 3; Indicates time The contextual importance of Temporal relationship to key events.

[0163] Step 4.2, dynamically adjust the temporal resolution and adaptively allocate the sampling frequency according to the importance of the scene;

[0164] The calculation formula is:

[0165] ;

[0166] in Indicates time The sampling frequency, Represents the result of scene importance assessment; function is a piecewise function, expressed as:

[0167] ;

[0168] in , , They represent high, medium and low time resolutions, corresponding to different sampling frequencies; and They are high and low threshold parameters, respectively, used to divide the importance levels.

[0169] Step 4.3, construct a hierarchical cache structure;

[0170] Hierarchical cache structure including high-resolution short-term cache and low-resolution long-term cache , the relationship between the two is:

[0171] : Save recent High-resolution image data over time for detailed analysis and abnormality diagnosis;

[0172] : Save recent Low-resolution or key-frame image data over time for long-term trend analysis;

[0173] ;

[0174] in and Respectively represent the time window size of short-term and long-term cache, Indicates much greater than;

[0175] Step 4.4, when an important event is detected, the backtracking analysis algorithm is triggered to temporarily increase the density of historical frame retention;

[0176] The calculation formula is:

[0177] ;

[0178] in and Represents the time points after and before adjustment, respectively. The historical frame retention density at is the number of image frames retained per unit time. is a time-weighted function used to weight the With events The correlation-adjusted preserving density can be defined as:

[0179] ;

[0180] in For events Time of occurrence; is the density improvement coefficient, which controls the maximum improvement range; is the time influence range parameter, which controls the decay rate of time influence; is an exponential function.

[0181] The output of this step is the adaptive time resolution and dynamically adjusted caching strategies, these outputs will directly affect the data collection and storage behavior of the system. By dynamically adjusting the sampling frequency according to the importance of the scene, the system is able to obtain higher quality data in critical scenes while saving resources in non-critical scenes. In addition, the design of the hierarchical cache structure and backtracking analysis algorithm ensures that the system can provide complete historical context information when abnormal events occur. These optimization strategies and caching decisions provide an operating framework and execution environment for the intelligent cache optimization in step 5, so that the final data management can achieve more refined value-oriented optimization on this basis.

[0182] Step 5: Based on the results of the abnormal propagation path probability matrix and the adaptive resource allocation system, an intelligent cache optimization algorithm is executed to achieve value-based data management and storage resource optimization allocation;

[0183] This step is based on all the previous steps, especially integrating the abnormal propagation path probability matrix of step 3 The adaptive resource allocation strategy in step 4 implements intelligent cache optimization, allocates more storage resources to targets that may be affected, and retains the most valuable information. This step is the final execution layer of the entire system, which converts the analysis results of all the previous steps into specific data management strategies. The specific implementation is as follows:

[0184] Step 5.1, optimizing storage resource allocation strategy;

[0185] The probability matrix of abnormal propagation paths predicted in step 3 , optimize storage resource allocation strategy:

[0186] ;

[0187] in Indicates the assignment to the node The amount of storage resources; Indicates an abnormal slave node Propagate to nodes The probability of , comes from the output of step 3; Representation Node The importance of can be evaluated based on factors such as target type and function; function It can be defined as:

[0188] ;

[0189] in represents the storage resource allocation function, The basic storage resource allocation is the minimum storage resource allocation for all targets; and They represent the impact coefficients of anomaly propagation probability and target importance on resource allocation respectively.

[0190] Step 5.2, construct an intelligent elimination algorithm based on information value to evaluate the retention value of each frame of image;

[0191] The calculation formula is:

[0192] ;

[0193] in Representing images The retention value score is used to determine the retention priority of the image when storage resources are limited; Representing images The amount of information can be measured by indicators such as information entropy or the number of key targets; Representing images The storage duration is the difference between the current time and the image acquisition time; Representing images Related to current events The correlation can be calculated by the matching degree between image content and event features;

[0194] function It can be defined as:

[0195] ;

[0196] in represents the image value scoring function after comprehensively considering the amount of information, storage time and event relevance; , , They represent the information volume, storage duration and relevance of the image to the event respectively; , , Respectively represent the weight coefficients of information volume, storage duration and event relevance; is the time decay factor, which controls the rate at which the image value decays over time; is an exponential function.

[0197] Step 5.3, combining storage resource constraints to implement an optimized elimination algorithm;

[0198] When cache space is insufficient, prioritize the elimination of value scores Lower the images, keep the high value images.

[0199] Step 5.4, implement the incremental learning model by continuously monitoring the system performance indicators;

[0200] System performance indicators such as key event capture rate, storage utilization, etc., dynamically adjust weight parameters , , , , Etc., and continue to optimize cache strategies.

[0201] Step 5.5, build a cache access index system;

[0202] It supports fast data retrieval based on multiple dimensions such as time, space, target type and event association, and improves the efficiency of accessing historical data.

[0203] The output of this step is a complete intelligent cache optimization system, which not only guides the system's data storage behavior, but also achieves self-optimization through incremental learning. In this way, the system can continuously improve the efficiency and accuracy of data management during long-term operation. At the same time, this step also forms a closed loop of the entire processing flow, because the optimized cache system will affect the acquisition and processing of future image data, which will become the input of step 1, thus forming a continuously optimized closed-loop system. Through the close connection and data flow of the five steps, this implementation method realizes an efficient, accurate, and resource-saving high-speed train image capture and processing system.

[0204] An AI-based high-speed train image capture processing system, used to execute the above-mentioned AI-based high-speed train image capture processing method, comprising:

[0205] Multi-dimensional information entropy analysis module, used to process high-speed train images based on a sliding window algorithm and identify key changes in scenes;

[0206] The dynamic association graph construction module is used to construct the spatial relationship network between targets based on the key change frames and generate the target association graph containing time series information;

[0207] The anomaly propagation prediction module is used to calculate the anomaly diffusion path between targets through multi-hop anomaly correlation analysis to form a risk probability matrix;

[0208] Adaptive resource allocation module, which is used to dynamically adjust monitoring parameters according to the importance of the scene and the risk of abnormality, and improve the data quality of key areas;

[0209] The intelligent cache management module is used to optimize storage strategies based on abnormal propagation risks and information value assessment to achieve value-oriented management of data.

[0210] Here, the present invention provides an implementation example:

[0211] The actual application test was carried out in a high-speed railway line monitoring system. The line is about 1,318 kilometers long, with about 680 monitoring points distributed along the line. Each point generates an average of about 8,640 images per day (one image is collected every 10 seconds). The following details the application scenario, implementation process, and technical effect verification.

[0212] This application selected the overhead line, track and bridge monitoring system of a certain section of a high-speed railway as the experimental platform. The section is about 150 kilometers long and has 72 image acquisition monitoring points, including 28 overhead line monitoring points, 32 track monitoring points, and 12 bridge structure monitoring points. The experiment lasted for 30 days and collected about 187 million original images with a total data volume of about 28TB.

[0213] In this application example, we performed multidimensional information entropy analysis on the image sequences collected at the same monitoring point at different times, setting the sliding window size N=20 and the weight coefficient =0.6, =0.4.

[0214] Under different weather and lighting conditions, the correspondence between the calculated scene change index and the actual abnormal events is shown in Table 1:

[0215] Table 1: Correspondence between scene change indicators and abnormal events under different conditions

[0216]

[0217] The multidimensional information entropy analysis algorithm can effectively identify abnormal scenes under various complex weather and lighting conditions, with an average accuracy of 88.3% and an average recall of 85.6%, showing good robustness.

[0218] In the example of overhead line monitoring, we use the dynamic association graph network model to model key components such as pillars, positioners, droppers, and contact wires. The correlation accuracy of the model at different sampling frequencies is compared, as shown in Table 2:

[0219] Table 2: Comparison of association accuracy of dynamic association graph network models at different sampling frequencies

[0220]

[0221] By comparison, it can be found that the dynamic association graph network model of this embodiment can improve the target association accuracy while maintaining low computing resource consumption under the condition of adaptive sampling frequency.

[0222] We conducted multi-hop anomaly correlation analysis on sleepers, fasteners, rails and other components in the track monitoring system. The accuracy of anomaly propagation prediction under different hop numbers is compared, as shown in Table 3:

[0223] Table 3: Anomaly propagation prediction results under different hop counts

[0224]

[0225] It can be seen that the multi-hop anomaly association analysis algorithm of this embodiment can effectively predict the propagation path of the anomaly, especially within the range of 1 hop and 2 hops, with the prediction accuracy exceeding 85%, and can warn of abnormal propagation events 8-24 minutes in advance, thus buying valuable intervention time for operation and maintenance personnel.

[0226] During the entire experiment, we compared the storage efficiency of the traditional fixed sampling method and the adaptive resource allocation method of this embodiment. The storage efficiency comparison results are shown in Table 4:

[0227] Table 4: Comparison of storage efficiency of different sampling strategies

[0228]

[0229] The data shows that the adaptive resource allocation algorithm of this implementation achieves a storage compression rate of 65%, while increasing the key event detection rate to 94.6%, which is far superior to traditional fixed sampling and simple threshold methods.

[0230] Through 30 consecutive days of application in a high-speed rail monitoring system, we have conducted a comprehensive evaluation of the long-term technical effect of this implementation method. The improvement effect of key performance indicators is shown in Table 5:

[0231] Table 5 Evaluation of long-term operation effect of this implementation method

[0232]

[0233] In practical applications, we have recorded multiple anomaly propagation prediction cases. The three most representative cases are shown in Table 6:

[0234] Table 6: Analysis of practical application cases of anomaly propagation prediction

[0235]

[0236] The comparison of monitoring system deployment costs under different implementation methods is shown in Table 7:

[0237] Table 7: Comparison of monitoring system deployment costs under different implementation methods

[0238]

[0239] This implementation method is not only superior to the traditional method in terms of technical indicators, but also has advantages in terms of economic benefits. By reducing storage and maintenance costs and improving fault prevention capabilities, the overall return on investment of the system has increased by 45.9 percentage points.

[0240] In practical applications, this implementation method improves the efficiency and reliability of the high-speed train image monitoring system, reduces the waste of storage resources, enhances the prediction and diagnosis capabilities of abnormal events, and provides strong technical support for the safe operation of high-speed railways.

[0241] The above describes an embodiment of the present invention, but this embodiment is not limited to the above-mentioned specific implementation mode. The above-mentioned specific implementation mode is merely illustrative and not restrictive. Under the guidance of this embodiment, ordinary technicians in this field can also make more forms of equivalent embodiments, all of which are within the protection of this embodiment.

Claims

1. A high-speed train image capture and processing method based on AI, characterized in that: include: The continuously acquired high-speed train images are processed based on a sliding window algorithm to identify key changes in the scene; Using the key changes in the scene, a dynamic association graph network model is constructed and a temporal graph attention network is applied to process the target spatial relationship and generate a target association graph; Based on the target association graph, a multi-hop anomaly association analysis algorithm is used to predict the propagation of anomalies between targets and obtain the probability matrix of anomaly propagation paths; Based on the key changes in the scene and the probability matrix of abnormal propagation paths, an adaptive resource allocation system is implemented to dynamically adjust image acquisition parameters according to the scene importance evaluation model; According to the results of the anomaly propagation path probability matrix and the adaptive resource allocation system, an intelligent cache optimization algorithm is executed to achieve value-based data management and storage resource optimization allocation.

2. According to the AI-based high-speed train image capture processing method of claim 1, it is characterized in that: The processing of the continuously acquired high-speed train images based on the sliding window algorithm includes: The sliding window algorithm is used to maintain the latest N frame image sequence; Calculate the information entropy difference between adjacent image frames; Calculate the structural similarity between adjacent image frames; The scene change index is constructed by combining information entropy difference and structural similarity.

3. The AI-based high-speed train image capture processing method according to claim 1 is characterized in that: The construction of the dynamic association graph network model includes: Use object detection algorithms to identify key sets of objects from images; Construct scene dynamic association graph; Calculate the association weights between targets; Applying temporal graph attention network to process dynamic correlations between objects.

4. The AI-based high-speed train image capture processing method according to claim 1 is characterized in that: The method of using a multi-hop anomaly association analysis algorithm to predict the propagation of anomalies between targets includes: Calculate the adjacency matrix and degree matrix corresponding to the target association graph; Construct a normalized adjacency matrix; Initialize the node’s abnormal attention vector; The attention propagation algorithm is used to iteratively calculate the k-hop anomaly associations; Construct an anomaly propagation model to predict the diffusion path of anomalies on the target graph.

5. The AI-based high-speed train image capture and processing method according to claim 1 is characterized in that: The adaptive resource allocation system comprises: Based on the scene change indicators, a scene importance assessment model is constructed; Dynamically adjust the temporal resolution and adaptively allocate sampling frequency according to the importance of the scene; Build a hierarchical cache structure; When an important event is detected, the retrospective analysis algorithm is triggered to temporarily increase the density of historical frame retention.

6. The AI-based high-speed train image capture and processing method according to claim 1 is characterized in that: The intelligent cache optimization algorithm includes: Optimize storage resource allocation strategy; Build an intelligent elimination algorithm based on information value to evaluate the retention value of each frame of image; Combined with storage resource constraints, an optimized elimination algorithm is implemented; Implementing incremental learning models by continuously monitoring system performance indicators; Build a cache access index system.

7. The AI-based high-speed train image capture processing method according to claim 2 is characterized in that: The calculation formula for constructing the scene change index is: ； in Indicates time The scene change indicator, and They represent the weight coefficients of information entropy difference and structural similarity in the scene change index, represents the information entropy difference between adjacent image frames, represents the structural similarity index, , Respectively indicate time ,time of the image.

8. The AI-based high-speed train image capture processing method according to claim 3 is characterized in that: The calculation formula for the dynamic association between the application time sequence graph attention network processing targets is: ； in is the scene dynamic association diagram, is the hidden state at the current moment, containing the feature representations of all nodes; is the observed feature at the current moment, that is, the target feature extracted from the image; represents the temporal graph attention network, which combines graph neural networks and attention algorithms to capture the correlation information between objects and maintain temporal coherence; is the updated hidden state, which contains temporal and spatial relationship information.

9. The AI-based high-speed train image capture processing method according to claim 4 is characterized in that: The calculation formula for constructing the anomaly propagation model and predicting the diffusion path of the anomaly on the target graph is: ； in Indicates that at the node Under abnormal conditions, the node The conditional probability of anomaly. Represents a slave node To Node The abnormal propagation attention value of and Respectively represent nodes and nodes Status characteristics; Calculates the conditional probability function.

10. A high-speed train image capture and processing system based on AI, characterized in that: A method for executing an AI-based high-speed train image capture and processing method as described in any one of claims 1 to 9, comprising: Multi-dimensional information entropy analysis module, used to process high-speed train images based on a sliding window algorithm and identify key changes in scenes; The dynamic association graph construction module is used to construct the spatial relationship network between targets based on the key change frames and generate the target association graph containing time series information; The anomaly propagation prediction module is used to calculate the anomaly diffusion path between targets through multi-hop anomaly correlation analysis to form a risk probability matrix; Adaptive resource allocation module, which is used to dynamically adjust monitoring parameters according to the importance of the scene and the risk of abnormality, and improve the data quality of key areas; The intelligent cache management module is used to optimize storage strategies based on abnormal propagation risks and information value assessment to achieve value-oriented management of data.

Images