A control method for intelligent inspection equipment for subway track line inspection

Abstract

The present invention relates to the field of patrol inspection equipment control technology, in particular to an intelligent patrol inspection equipment control method for subway track line patrol inspection, which obtains the geographical location information of each subway track area to be patrolled, generates a patrol route for the patrol inspection equipment according to the geographical location information of each subway track area to be patrolled; controls the patrol inspection equipment to patrol the subway track to be patrolled according to the patrol route; obtains the subway track feature image of the subway track area to be patrolled through the camera mechanism carried by the patrol inspection equipment, and performs feature extraction on the subway track feature image; analyzes the subway track area to be patrolled according to the LBP feature matrix of the subway track area to be patrolled, and generates a patrol inspection result. The present invention improves the patrol inspection and dispatching capabilities of the patrol inspection equipment, effectively saves patrol inspection resources, improves patrol inspection efficiency and accuracy, and ensures the safety and normal operation of the subway track.

Description

Technical Field

[0001] The present invention relates to the technical field of patrol inspection equipment control, and in particular to a method for controlling intelligent patrol inspection equipment for patrol inspection of subway track lines. Background Art

[0002] With the acceleration of urbanization, subways, as the core carrier of urban rail transit, face severe challenges in their safe operation. The degradation of rail lines is significantly affected by the dynamic coupling of environmental factors. Traditional inspections mostly use fixed-cycle manual inspections or single sensor monitoring, which have problems such as response lag, resource waste and low risk identification rate. In the existing technology, although there are studies on the application of Internet of Things technology to rail monitoring, they mostly focus on the collection of static environmental parameters and lack the ability to dynamically model the synergistic effects of multiple factors such as temperature and humidity, geological deformation, and electromagnetic interference, resulting in insufficient degradation prediction accuracy. Especially under the trend of subway network complexity, conventional inspection route planning algorithms are difficult to take into account real-time risk distribution and equipment scheduling efficiency. They often use full-line coverage inspections, resulting in many unnecessary inspection operations and serious waste of inspection equipment resources. In addition, rail surface defect detection technology based on visible light images generally has defects such as single feature extraction dimension and poor texture analysis robustness. Traditional algorithms such as LBP (local binary pattern) have a high false detection rate under dynamic lighting conditions, resulting in low reliability of inspection results obtained by inspection equipment. Summary of the invention

[0003] The present invention overcomes the shortcomings of the prior art and provides a control method for intelligent inspection equipment used for subway track line inspection.

[0004] To achieve the above-mentioned purpose, the technical solution adopted by the present invention is: The present invention discloses a control method for intelligent inspection equipment for subway track line inspection, comprising the following steps: Obtaining the degradation index of the subway track to be inspected under the combined effect of various environmental factor data, and constructing a degradation index hash dictionary of the subway track to be inspected according to the degradation index of the subway track to be inspected under the combined effect of various environmental factor data; The subway track to be inspected is divided into several sub-subway track areas, and a historical dynamic environmental factor data set of each sub-subway track area within a preset time period is obtained. According to the historical dynamic environmental factor data set and in combination with a degradation index hash dictionary, a degradation index evolution curve of each sub-subway track area attenuated and aggregated over time is constructed; Each sub-subway track area is evaluated according to the degradation index evolution curve of each sub-subway track area attenuated over time, and the dynamic risk score value of each sub-subway track area is obtained; the sub-subway track area that needs to be inspected and the sub-subway track area that does not need to be inspected are determined according to the dynamic risk score value of each sub-subway track area; Obtaining geographical location information of each sub-subway track area to be inspected, generating an inspection route for the inspection equipment according to the geographical location information of each sub-subway track area to be inspected; and controlling the inspection equipment to inspect the subway track to be inspected according to the inspection route; When the inspection device moves to the sub-subway track area to be inspected, a subway track feature image of the sub-subway track area to be inspected is obtained through a camera mechanism carried by the inspection device, and features are extracted from the subway track feature image to obtain an LBP feature matrix of the sub-subway track area to be inspected; The sub-subway track area to be inspected is analyzed according to the LBP feature matrix of the sub-subway track area to be inspected, an inspection result is generated, and the inspection result is transmitted to the memory of the inspection device for storage.

[0005] Preferably, the degradation index of the subway track to be inspected under the combined effect of various environmental factor data is obtained, and a degradation index hash dictionary of the subway track to be inspected is constructed according to the degradation index of the subway track to be inspected under the combined effect of various environmental factor data, specifically: Obtaining rail material information of the subway track to be inspected, searching the degradation index of the subway track to be inspected under the combined effect of various environmental factor data in the big data network according to the rail material information; and constructing a dynamic hash mapping table; Discretize each environmental factor data combination to obtain the factor discrete value of each environmental factor data combination; and combine each factor discrete value into a binary feature code according to a preset weight to generate a unique hash identification key for each environmental factor data combination; Generate a number of key-value pairs in the dynamic hash mapping table according to the hash identification key corresponding to each environmental factor data combination; associate the degradation index under the action of each environmental factor data combination with the corresponding hash identification key to perform key-value pair mapping to obtain the linked list storage slot mapped by each hash identification key; The corresponding degradation indexes are written into the linked list storage slots mapped by the corresponding hash identification keys in the dynamic hash mapping table through the double hash conflict resolution mechanism, and the degradation index hash dictionary of the subway track to be inspected is obtained.

[0006] Preferably, the subway track to be inspected is divided into several sub-subway track areas, and a historical dynamic environmental factor data set of each sub-subway track area within a preset time period is obtained. According to the historical dynamic environmental factor data set and in combination with the degradation index hash dictionary, a degradation index evolution curve of each sub-subway track area attenuated and aggregated over time is constructed, specifically: Perform time series feature analysis on the historical dynamic environmental factor data set of each sub-subway track area within a preset time period, divide the dynamic timestamps based on the sliding window mechanism, extract the corresponding historical dynamic environmental factor data at each timestamp, and obtain the historical dynamic environmental factor data of each sub-subway track area at each timestamp; Convert the historical dynamic environmental factor data of each sub-subway track area at each time stamp into a corresponding binary feature code to obtain a hash identification key corresponding to the historical dynamic environmental factor data of each sub-subway track area at each time stamp; Importing the hash identification key corresponding to the historical dynamic environmental factor data of each sub-subway track area at each time stamp into the degradation index hash dictionary; In order of timestamps, the similarities between the hash identification key corresponding to the historical dynamic environmental factor data of the corresponding timestamp and the hash identification key stored in each key-value pair in the degradation index hash dictionary are calculated in sequence, and the key-value pair with the highest similarity is marked, and the degradation index of the sub-subway track area at the corresponding timestamp is extracted from the linked list storage slot of the marked key-value pair; and the degradation index of each sub-subway track area based on the timestamp order is obtained by analogy; An exponential decay function is preset, and based on the exponential decay function, the degradation index of each sub-subway track area based on the timestamp order is time-weighted, and an evolution curve of the degradation index of each sub-subway track area that decays and aggregates over time is output.

[0007] Preferably, each sub-subway track area is evaluated according to the degradation index evolution curve of each sub-subway track area attenuated over time, and the dynamic risk score value of each sub-subway track area is obtained, which is specifically: Perform time window sliding segmentation on the degradation index evolution curve of each sub-subway track area to extract local time series segments; calculate the ratio of the standard deviation to the mean of the local time series segments to obtain the coefficient of variation of each sub-subway track area; and calculate the first-order derivative and the second-order derivative of the degradation index evolution curve of each sub-subway track area; The coefficient of variation, first-order derivative and second-order derivative of the degradation curve of each sub-region are synchronized by time stamp through the multi-source data spatiotemporal alignment algorithm to obtain the synchronized coefficient of variation, synchronized first-order derivative and synchronized second-order derivative under the same time reference; Deploy a multi-dimensional feature fusion channel and use spatiotemporal convolution kernels to extract the fluctuation frequency domain characteristics of the synchronized coefficient of variation, the trend directional characteristics of the synchronized first-order derivative, and the degradation acceleration characteristics of the synchronized second-order derivative. A multi-head attention mechanism is introduced to perform cross-dimensional interaction on the fluctuation frequency domain characteristics of the synchronized coefficient of variation, the trend directional characteristics of the synchronized first-order derivative, and the degradation acceleration characteristics of the synchronized second-order derivative, generating a fused feature tensor containing spatiotemporal correlation weights. The fused feature tensor is input into the deep feature fusion network, and the multi-scale spatiotemporal correlation pattern is extracted using residual connection and feature pyramid structure, and a three-dimensional risk indicator vector including volatility intensity, trend stability and acceleration probability is output; The three-dimensional risk index vector is mapped to a preset risk level space, and the discrete levels are nonlinearly weighted aggregated through a radial basis function neural network to generate a dynamic risk score value for each sub-subway track area.

[0008] Preferably, the sub-subway track areas that need to be inspected and the sub-subway track areas that do not need to be inspected are determined according to the dynamic risk score value of each sub-subway track area, specifically: Obtaining a dynamic risk score value of each sub-subway track area and setting a dynamic risk score value threshold; determining whether the dynamic risk score value of each sub-subway track area is greater than the dynamic risk score value threshold; If the dynamic risk score value of a certain sub-subway track area is greater than the dynamic risk score value threshold, the sub-subway track area is determined as a sub-subway track area that needs to be inspected; If the dynamic risk score value of a certain sub-subway track area is not greater than the dynamic risk score value threshold, the sub-subway track area is determined as a sub-subway track area that does not require inspection.

[0009] Preferably, the geographical location information of each sub-subway track area to be inspected is obtained, and the inspection route of the inspection equipment is generated according to the geographical location information of each sub-subway track area to be inspected, specifically: Obtain a track distribution map of the subway track to be inspected, and mark the geographical location information of each sub-subway track area to be inspected in the track distribution map; The particle swarm optimization algorithm is introduced to construct the initial solution space based on the geographical location information of each subway track area to be inspected. Each sub-area is abstracted as a particle position in the particle swarm optimization algorithm, and the speed and position distribution of the particle swarm are initialized according to the preset moving speed and starting position of the inspection equipment. By continuously distributing the speed and position of the current particle group, each particle is moved and updated in the solution space according to the preset rules. After a preset number of iterations, several different moving paths of particles from the starting position to each sub-subway track area to be inspected are obtained; The total path length value of each moving path is calculated; and the moving path with the shortest total path length value is used as the inspection route of the inspection equipment.

[0010] Preferably, feature extraction is performed on the subway track feature image to obtain the LBP feature matrix of the sub-subway track area to be inspected, specifically: The collected subway track images are preprocessed by grayscale conversion and illumination balance to enhance local texture contrast; a dynamic radius strategy is used to generate annular sampling points within the pixel neighborhood; A multi-scale differential comparison mechanism is introduced to perform secondary threshold quantization on the grayscale difference between the central pixel and the ring sampling point to generate a rotationally invariant binary code sequence. The local texture descriptor is constructed through a sliding window, and the direction is weighted and fused in combination with the track direction to eliminate the feature offset caused by mechanical vibration. The whole image is scanned in blocks and overlapped, and the multi-level LBP response images are integrated and normalized dimension reduction is performed to generate an LBP feature matrix with spatial distribution characteristics.

[0011] Preferably, the sub-subway track area to be inspected is analyzed according to the LBP feature matrix of the sub-subway track area to be inspected, an inspection result is generated, and the inspection result is transmitted to the memory of the inspection device for storage, specifically: Obtaining LBP feature matrices corresponding to various early warning defects of subway tracks through a big data network, building a database, and importing the LBP feature matrices corresponding to various early warning defects of subway tracks into the database; After obtaining the LBP feature matrix of a certain subway track area that needs to be inspected, importing the LBP feature matrix of the subway track area that needs to be inspected into the database; Calculate the matrix overlap between the LBP feature matrix of the sub-subway track area to be inspected and each LBP feature matrix in the database; If the matrix overlap between the LBP feature matrix of the sub-subway track area to be inspected and each LBP feature matrix in the database is not greater than the preset overlap threshold, it means that there is no warning defect in the sub-subway track area to be inspected, and the sub-subway track area to be inspected is marked as a subway track area that does not require maintenance; If the matrix overlap between the LBP feature matrix of the sub-subway track area to be inspected and one or more LBP feature matrices in the database is greater than a preset overlap threshold, it means that there is a warning defect in the sub-subway track area to be inspected, and the sub-subway track area to be inspected is marked as a subway track area that needs maintenance; An inspection result is generated according to the subway track area that needs maintenance and the subway track area that does not need maintenance, and the inspection result is transmitted to a memory of the inspection device for storage.

[0012] The following steps are also included: If a certain sub-subway track area requiring inspection is marked as a subway track area requiring maintenance, obtaining geographic location information of the subway track area requiring maintenance; At the same time, the inspection equipment is controlled to obtain the topographic image information of the subway track area that needs maintenance within a preset radius; and the warning defect type of the subway track area that needs maintenance; The geographical location information, topographic image information and warning defect type of the subway track area requiring maintenance are compressed and bundled to generate a subway track maintenance information data packet, and the subway track maintenance information data packet is transmitted to the memory of the inspection equipment for storage.

[0013] The present invention solves the technical defects existing in the background technology, and has the following beneficial effects: constructing a degradation index evolution curve of each sub-subway track area that decays and aggregates over time, and evaluating each sub-subway track area according to the degradation index evolution curve of each sub-subway track area that decays and aggregates over time; determining the sub-subway track area that needs to be inspected and the sub-subway track area that does not need to be inspected according to the dynamic risk score value of each sub-subway track area; obtaining the geographical location information of each sub-subway track area that needs to be inspected, and generating an inspection route of the inspection equipment according to the geographical location information of each sub-subway track area that needs to be inspected; controlling the inspection equipment to inspect the subway track to be inspected according to the inspection route; when the inspection equipment moves to the sub-subway track area that needs to be inspected, obtaining the subway track feature image of the sub-subway track area that needs to be inspected through the camera mechanism carried by the inspection equipment, extracting features from the subway track feature image, and obtaining the LBP feature matrix of the sub-subway track area that needs to be inspected; analyzing the sub-subway track area that needs to be inspected according to the LBP feature matrix of the sub-subway track area that needs to be inspected, and generating an inspection result. The present invention improves the inspection and dispatching capabilities of inspection equipment, effectively saves inspection resources, improves inspection efficiency and accuracy, and ensures the safety and normal operation of subway tracks. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, drawings of other embodiments can be obtained based on these drawings without paying creative work.

[0015] Figure 1 This is a flow chart of the first method of the intelligent inspection equipment control method; Figure 2 This is a flow chart of the second method of the intelligent inspection equipment control method; Figure 3 This is a flow chart of the third method of the intelligent inspection equipment control method. DETAILED DESCRIPTION

[0016] In order to more clearly understand the above-mentioned purpose, features and advantages of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be noted that the embodiments of the present application and the features in the embodiments can be combined with each other without conflict.

[0017] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Therefore, the protection scope of the present invention is not limited to the specific embodiments disclosed below.

[0018] like Figure 1 As shown, the present invention discloses a control method of intelligent inspection equipment for subway track line inspection, comprising the following steps: S102, obtaining a degradation index of the subway track to be inspected under the combined effects of various environmental factor data, and constructing a degradation index hash dictionary of the subway track to be inspected according to the degradation index of the subway track to be inspected under the combined effects of various environmental factor data; S104, dividing the subway track to be inspected into several sub-subway track areas, obtaining a historical dynamic environmental factor data set of each sub-subway track area within a preset time period, and constructing a degradation index evolution curve of each sub-subway track area that decays and aggregates over time based on the historical dynamic environmental factor data set and in combination with a degradation index hash dictionary; wherein the historical dynamic environmental factor data set of each sub-subway track area within the preset time period can be obtained from the monitoring log of the subway track to be inspected; S106, evaluating each sub-subway track area according to the degradation index evolution curve of each sub-subway track area attenuated over time, and obtaining a dynamic risk score value of each sub-subway track area; determining the sub-subway track area that needs to be inspected and the sub-subway track area that does not need to be inspected according to the dynamic risk score value of each sub-subway track area; S108, obtaining geographical location information of each sub-subway track area to be inspected, generating an inspection route for the inspection equipment according to the geographical location information of each sub-subway track area to be inspected; and controlling the inspection equipment to inspect the subway track to be inspected according to the inspection route; S110, when the inspection device moves to the sub-subway track area to be inspected, a subway track feature image of the sub-subway track area to be inspected is obtained by a camera mechanism carried by the inspection device, and features are extracted from the subway track feature image to obtain an LBP feature matrix of the sub-subway track area to be inspected; S112, analyzing the sub-subway track area to be inspected according to the LBP feature matrix of the sub-subway track area to be inspected, generating an inspection result, and transmitting the inspection result to a memory of an inspection device for storage.

[0019] Preferably, the degradation index of the subway track to be inspected under the combined effect of various environmental factor data is obtained, and a degradation index hash dictionary of the subway track to be inspected is constructed according to the degradation index of the subway track to be inspected under the combined effect of various environmental factor data, such as Figure 2 As shown, specifically: S202, obtaining rail material information of the subway track to be inspected, searching the degradation index of the subway track to be inspected under the combined effect of various environmental factor data in the big data network according to the rail material information; and constructing a dynamic hash mapping table; The environmental factors include, but are not limited to, temperature and humidity, geological deformation, electromagnetic interference, rainfall, salt spray concentration, and traffic load; S204, discretizing each environmental factor data combination to obtain a factor discrete value of each environmental factor data combination; and combining each factor discrete value into a binary feature code according to a preset weight to generate a unique hash identification key for each environmental factor data combination; S206, generating a plurality of key-value pairs in the dynamic hash mapping table according to the hash identification keys corresponding to the environmental factor data combinations; performing key-value pair association mapping between the degradation index under the action of the environmental factor data combinations and the corresponding hash identification keys, and obtaining linked list storage slots mapped by the hash identification keys; S208. Write the corresponding degradation indexes into the linked list storage slots mapped by the corresponding hash identification keys in the dynamic hash mapping table through a double hash conflict resolution mechanism to obtain a degradation index hash dictionary of the subway track to be inspected.

[0020] The rail material information of the subway track to be inspected refers to the collection of various characteristics and related data about the rails used in the subway track to be inspected. It includes the specific material of the rails, such as what kind of steel or alloy it is; it also covers information on the physical properties of the rails, such as hardness, strength, toughness and other indicators; it may also involve details such as the chemical composition of the rails.

[0021] The deterioration index is a quantitative indicator used to measure and evaluate the degree of gradual deterioration of the performance or status of the subway track to be inspected. It comprehensively reflects the decline of the quality, reliability, safety and other aspects of the subway track relative to the initial state under the influence of various factors, such as environmental conditions (temperature and humidity, geological deformation, etc.), service life, traffic load, etc. The deterioration index can intuitively indicate the current deterioration level of the subway track. The higher the value, the more serious the deterioration of the track, and there may be more potential problems or risks.

[0022] It should be noted that the rail material information is first obtained in order to accurately retrieve the relevant degradation index in the big data network, and at the same time a dynamic hash mapping table is constructed to establish an infrastructure for subsequent data processing and storage. Discretizing the environmental factor data combination and generating a unique hash identification key helps to effectively identify and classify complex environmental factor combinations. By generating key-value pairs and performing associative mapping, a corresponding relationship between the environmental factor data combination and the degradation index is established. The degradation index is written into the corresponding storage slot using a double hash conflict resolution mechanism to ensure the accuracy and efficiency of data storage. The present invention realizes efficient organization, classification, storage and association of these complex data by constructing a degradation index hash dictionary, so that the corresponding degradation index can be quickly and accurately found according to the environmental factor combination, providing a convenient query method for monitoring, evaluation and maintenance of subway tracks, and improving inspection efficiency.

[0023] Preferably, the subway track to be inspected is divided into several sub-subway track areas, and a historical dynamic environmental factor data set of each sub-subway track area within a preset time period is obtained. According to the historical dynamic environmental factor data set and in combination with the degradation index hash dictionary, a degradation index evolution curve of each sub-subway track area attenuated over time is constructed, such as Figure 3 As shown, specifically: S302, performing time series feature analysis on the historical dynamic environmental factor data set of each sub-subway track area within a preset time period, dividing the dynamic timestamps based on a sliding window mechanism, extracting the corresponding historical dynamic environmental factor data at each timestamp, and obtaining the historical dynamic environmental factor data of each sub-subway track area at each timestamp; S304, converting the historical dynamic environmental factor data of each sub-subway track area at each time stamp into a corresponding binary feature code, and obtaining a hash identification key corresponding to the historical dynamic environmental factor data of each sub-subway track area at each time stamp; S306, importing the hash identification key corresponding to the historical dynamic environmental factor data of each sub-subway track area at each time stamp into the degradation index hash dictionary; S308, in accordance with the order of timestamps, sequentially calculate the similarity between the hash identification key corresponding to the historical dynamic environmental factor data of the corresponding timestamp and the hash identification key stored in each key-value pair in the degradation index hash dictionary, mark the key-value pair with the highest similarity, and extract the degradation index of the sub-subway track area at the corresponding timestamp from the linked list storage slot of the marked key-value pair; and so on, obtain the degradation index of each sub-subway track area based on the timestamp order; S310 , preset an exponential decay function, perform time-effectiveness weighting on the degradation index of each sub-subway track area based on the timestamp sequence based on the exponential decay function, and output a degradation index evolution curve of each sub-subway track area that decays and aggregates over time.

[0024] Among them, an exponential decay function is preset, and the degradation index of each sub-subway track area based on the timestamp sequence is weighted based on the exponential decay function, and the degradation index evolution curve of each sub-subway track area decayed over time is output, specifically: the specific form and parameters of the exponential decay function are set by relevant technical personnel. Then, each sub-subway track area is traversed. For each sub-subway track area, a degradation index sequence based on the timestamp sequence is obtained. Starting from the earliest timestamp, the degradation index corresponding to the current timestamp is multiplied by the decay weight calculated according to the exponential decay function to obtain the weighted degradation index value. Then, this weighted value is accumulated into a cumulative variable. Then move to the next timestamp and repeat the calculation and accumulation operation of the degradation index and the decay weight. In this process, the cumulative variable is continuously updated. After traversing all the timestamps of the sub-subway track area, the value of the cumulative variable obtained is the degradation index of the sub-subway track area after decaying over time. Repeat this operation for all sub-subway track areas, so as to obtain the degradation index of each sub-subway track area decayed over time. Finally, based on these data, the degradation index evolution curve of each sub-subway track area with attenuation aggregation over time is drawn to intuitively show the trend and degree of change of the degradation index of each sub-subway track area over time.

[0025] It should be noted that the subway tracks to be inspected are subdivided to more accurately analyze the situation in each area. The historical dynamic environmental factor data set of each sub-subway track area is parsed and processed, and the key features are extracted and converted into hash identification keys, so that they can be linked with the degradation index hash dictionary. The corresponding degradation index is found by calculating the similarity, and the association from environmental factor data to degradation index is achieved. Then, the degradation index is weighted to reflect the influence of time factors on degradation, and finally the degradation index evolution curve is generated. In general, by dividing the subway track into regions, obtaining the historical environmental factor data of each region and conducting in-depth processing, combined with the degradation index hash dictionary, the degradation index of each sub-subway track area at different time points is accurately calculated, and the degradation index evolution curve with time decay aggregation is generated by timeliness weighting, so as to quantify the trend and degree of degradation of each sub-subway track area over time.

[0026] Preferably, each sub-subway track area is evaluated according to the degradation index evolution curve of each sub-subway track area attenuated over time, and the dynamic risk score value of each sub-subway track area is obtained, which is specifically: Perform time window sliding segmentation on the degradation index evolution curve of each sub-subway track area to extract local time series segments; calculate the ratio of the standard deviation to the mean of the local time series segments to obtain the coefficient of variation of each sub-subway track area; and calculate the first-order derivative and second-order derivative of the degradation index evolution curve of each sub-subway track area; wherein the first-order derivative is the slope of the degradation index evolution curve, and the second-order derivative is the curvature of the degradation index evolution curve; It should be noted that the time window sliding segmentation and coefficient of variation calculation quantify local volatility through the standard deviation / mean ratio, eliminate absolute dimension interference, and reveal the relative abnormal amplitude of the degradation process; the first-order derivative (slope) captures the linear trend of the degradation rate, and the second-order derivative (curvature) identifies the nonlinear mutation of the degradation acceleration. The combination of the two can distinguish between gradual degradation and sudden deterioration.

[0027] The coefficient of variation, first-order derivative and second-order derivative of the degradation curve of each sub-region are synchronized by time stamp through the multi-source data spatiotemporal alignment algorithm to obtain the synchronized coefficient of variation, synchronized first-order derivative and synchronized second-order derivative under the same time reference; It should be noted that the spatiotemporal alignment algorithm solves the timing deviation problem of multi-source heterogeneous data (such as sensor data, inspection records, and environmental parameters) through timestamp synchronization processing, and establishes the basis for spatiotemporal consistency analysis.

[0028] Deploy a multi-dimensional feature fusion channel and use spatiotemporal convolution kernels to extract the fluctuation frequency domain characteristics of the synchronized coefficient of variation, the trend directional characteristics of the synchronized first-order derivative, and the degradation acceleration characteristics of the synchronized second-order derivative. A multi-head attention mechanism is introduced to perform cross-dimensional interaction on the fluctuation frequency domain characteristics of the synchronized coefficient of variation, the trend directional characteristics of the synchronized first-order derivative, and the degradation acceleration characteristics of the synchronized second-order derivative, generating a fused feature tensor containing spatiotemporal correlation weights. It should be noted that the spatiotemporal convolution kernel uses a three-dimensional convolution structure (time axis + space axis + feature axis) to extract: fluctuation frequency domain features (identifying periodic anomalies through Fourier transform), trend directional features (sliding average filter to enhance trend signals), degradation acceleration characteristics (quadratic difference enhancement mutation response). The multi-head attention mechanism establishes a three-dimensional correlation matrix of fluctuation intensity-trend direction-acceleration through parallel multi-group self-attention calculations to capture cross-dimensional coupling effects (such as high-frequency fluctuations triggering acceleration threshold transitions).

[0029] The fused feature tensor is input into the deep feature fusion network, and the multi-scale spatiotemporal correlation pattern is extracted using residual connection and feature pyramid structure, and a three-dimensional risk indicator vector including volatility intensity, trend stability and acceleration probability is output; It should be noted that residual connections retain shallow local features (such as single-point anomalies), and feature pyramids fuse multi-scale spatiotemporal patterns (short-term pulse interference and long-term trend drift) to build a "micro-meso-macro" risk identification system.

[0030] The three-dimensional risk index vector is mapped to a preset risk level space, and the discrete levels are nonlinearly weighted aggregated through a radial basis function neural network to generate a dynamic risk score value for each sub-subway track area.

[0031] It should be noted that the radial basis function neural network realizes nonlinear mapping of the risk level space through the Gaussian kernel function. Its local response characteristics can finely distinguish the transition intervals between adjacent risk levels and avoid the boundary blur caused by traditional linear weighting.

[0032] In general, this step can comprehensively quantify the real-time risk status of different sub-subway track areas, reflecting both the instantaneous intensity of degradation (such as mutation risk) and its long-term evolution law (such as trend stability), providing a high-precision risk assessment basis for subway track maintenance, supporting the upgrade of decision-making mode from "passive response" to "active prediction", effectively reducing the probability of sudden failures and optimizing resource allocation, helping to timely discover high-risk areas and take targeted measures to ensure the safe and stable operation of the subway track system, and improving the accuracy, comprehensiveness and timeliness of risk assessment of subway track areas.

[0033] Preferably, the sub-subway track areas that need to be inspected and the sub-subway track areas that do not need to be inspected are determined according to the dynamic risk score value of each sub-subway track area, specifically: Obtaining a dynamic risk score value of each sub-subway track area and setting a dynamic risk score value threshold; determining whether the dynamic risk score value of each sub-subway track area is greater than the dynamic risk score value threshold; If the dynamic risk score value of a certain sub-subway track area is greater than the dynamic risk score value threshold, the sub-subway track area is determined as a sub-subway track area that needs to be inspected; If the dynamic risk score value of a certain sub-subway track area is not greater than the dynamic risk score value threshold, the sub-subway track area is determined as a sub-subway track area that does not require inspection.

[0034] It should be noted that by obtaining the dynamic risk score of each sub-subway track area and comparing it with the set threshold, the sub-subway track area can be quickly and accurately divided into two categories: those that need to be inspected and those that do not need to be inspected. This can make the inspection of the subway track more targeted, improve the efficiency of resource utilization, and timely focus on and inspect the high-risk sub-subway track areas that need to be inspected, which helps to discover potential problems in advance and take corresponding measures.

[0035] Preferably, the geographical location information of each sub-subway track area to be inspected is obtained, and the inspection route of the inspection equipment is generated according to the geographical location information of each sub-subway track area to be inspected, specifically: Obtain a track distribution map of the subway track to be inspected, and mark the geographical location information of each sub-subway track area to be inspected in the track distribution map; The particle swarm optimization algorithm is introduced to construct the initial solution space based on the geographical location information of each subway track area to be inspected. Each sub-area is abstracted as a particle position in the particle swarm optimization algorithm, and the speed and position distribution of the particle swarm are initialized according to the preset moving speed and starting position of the inspection equipment. By continuously distributing the speed and position of the current particle group, each particle is moved and updated in the solution space according to the preset rules. After a preset number of iterations, several different moving paths of particles from the starting position to each sub-subway track area to be inspected are obtained; The total path length value of each moving path is calculated; and the moving path with the shortest total path length value is used as the inspection route of the inspection equipment.

[0036] It should be noted that by obtaining the geographical location information of the sub-subway track area that needs to be inspected and marking it in the track distribution map, combined with the particle swarm optimization algorithm, the optimal inspection route can be efficiently generated. This can not only ensure that the inspection equipment covers all the sub-subway track areas that need to be inspected, but also improve the inspection efficiency and reduce energy consumption.

[0037] Preferably, feature extraction is performed on the subway track feature image to obtain the LBP feature matrix of the sub-subway track area to be inspected, specifically: The collected subway track images are preprocessed by grayscale conversion and illumination balance to enhance local texture contrast; a dynamic radius strategy is used to generate annular sampling points within the pixel neighborhood; A multi-scale differential comparison mechanism is introduced to perform secondary threshold quantization on the grayscale difference between the central pixel and the ring sampling point to generate a rotationally invariant binary code sequence. Among them, a multi-scale differential comparison mechanism is introduced, and the grayscale difference between the central pixel and the annular sampling point is subjected to secondary threshold quantization. The specific steps of generating a rotation-invariant binary coding sequence are as follows: first, the position of the central pixel and the surrounding annular sampling points is determined, and the grayscale difference between the central pixel and each annular sampling point is calculated; secondly, these grayscale differences are threshold quantized once and converted into preliminary binary codes; then, different secondary threshold quantizations are applied to annular sampling points of different scales to further refine the binary codes; finally, the rotation-invariant coding rules are used to ensure that the generated binary sequence is not affected by image rotation, thereby obtaining a stable feature representation.

[0038] The local texture descriptor is constructed through a sliding window, and the direction is weighted and fused in combination with the track direction to eliminate the feature offset caused by mechanical vibration. The specific steps of constructing local texture descriptors through sliding windows and performing directional weighted fusion in combination with the track direction to eliminate feature offset caused by mechanical vibration are as follows: define a sliding window of fixed size, slide it pixel by pixel on the subway track image, and extract local texture information in the window after each slide. In each sliding window, calculate local texture features (such as gray-level co-occurrence matrix, gradient histogram, etc.) to generate local texture descriptors; detect the direction of the track through image processing technology (such as edge detection, Hough transform, etc.) to determine the direction information of the track; assign a directional weight to each local texture descriptor according to the direction of the track; fuse the local texture descriptors with directional weights to generate a global texture descriptor.

[0039] The whole image is scanned in blocks and overlapped, multi-level LBP response images are integrated and normalized dimension reduction is performed to generate an LBP feature matrix with spatial distribution characteristics; Among them, the specific steps of performing overlapping block scanning on the entire image, integrating multi-level LBP response maps and performing normalized dimensionality reduction processing to generate an LBP feature matrix with spatial distribution characteristics are as follows: first, the entire image is divided into multiple overlapping small blocks, and the LBP feature response is calculated for each small block; then, the multi-level LBP response maps of different scales and directions are integrated to form a complete feature map; then, the feature map is normalized to ensure that the eigenvalues ​​are on a uniform scale; finally, the high-dimensional feature map is compressed into a low-dimensional feature matrix through dimensionality reduction technology (such as principal component analysis) to generate an LBP feature matrix with spatial distribution characteristics, thereby retaining key texture information and reducing data redundancy.

[0040] In summary, the above steps not only enhance the robustness and stability of features, reduce the interference of factors such as light and mechanical vibration, but also improve the accuracy and comprehensiveness of feature extraction, provide high-quality feature description for intelligent inspection of subway tracks, help to more accurately identify and locate track defects, and improve the level of intelligence in subway track maintenance.

[0041] Preferably, the sub-subway track area to be inspected is analyzed according to the LBP feature matrix of the sub-subway track area to be inspected, an inspection result is generated, and the inspection result is transmitted to the memory of the inspection device for storage, specifically: Obtaining LBP feature matrices corresponding to various early warning defects of subway tracks through a big data network, building a database, and importing LBP feature matrices corresponding to various early warning defects of subway tracks into the database; wherein the early warning defects include but are not limited to cracks, wear, deformation, rust, looseness, foreign matter and uneven settlement; After obtaining the LBP feature matrix of a certain subway track area that needs to be inspected, importing the LBP feature matrix of the subway track area that needs to be inspected into the database; Calculate the matrix overlap between the LBP feature matrix of the sub-subway track area to be inspected and each LBP feature matrix in the database; If the matrix overlap between the LBP feature matrix of the sub-subway track area to be inspected and each LBP feature matrix in the database is not greater than the preset overlap threshold, it means that there is no warning defect in the sub-subway track area to be inspected, and the sub-subway track area to be inspected is marked as a subway track area that does not require maintenance; If the matrix overlap between the LBP feature matrix of the sub-subway track area to be inspected and one or more LBP feature matrices in the database is greater than a preset overlap threshold, it means that there is a warning defect in the sub-subway track area to be inspected, and the sub-subway track area to be inspected is marked as a subway track area that needs maintenance; An inspection result is generated according to the subway track area that needs maintenance and the subway track area that does not need maintenance, and the inspection result is transmitted to a memory of the inspection device for storage.

[0042] The following steps are also included: If a certain sub-subway track area requiring inspection is marked as a subway track area requiring maintenance, obtaining geographic location information of the subway track area requiring maintenance; At the same time, the inspection equipment is controlled to obtain the topographic image information of the subway track area that needs maintenance within a preset radius; and the warning defect type of the subway track area that needs maintenance; If a certain subway track area that needs to be inspected is marked as a subway track area that needs maintenance, the specific warning defect type is determined by the LBP feature matrix of the area and the LBP feature matrix matched in the database with a matrix overlap greater than a preset overlap threshold, such as cracks, wear, deformation, rust, looseness, foreign matter or uneven settlement; The geographical location information, topographic image information and warning defect type of the subway track area requiring maintenance are compressed and bundled to generate a subway track maintenance information data packet, and the subway track maintenance information data packet is transmitted to the memory of the inspection equipment for storage.

[0043] It should be noted that the LBP feature matrix corresponding to various warning defects of the subway track is obtained through the big data network, and a database is constructed. The obtained LBP feature matrix of the sub-subway track area that needs to be inspected is imported into the database for feature matching and defect detection. The overlap between the LBP feature matrix of the sub-subway track area that needs to be inspected and each LBP feature matrix in the database is calculated to determine whether there is a warning defect. If the overlap between the LBP feature matrix of the sub-subway track area that needs to be inspected and each LBP feature matrix in the database is not greater than the preset threshold, it is considered that there is no warning defect in the area; conversely, if the overlap is greater than the preset threshold, it is considered that there is a warning defect in the area and it is marked accordingly. The inspection results are generated according to the subway track areas that need maintenance and those that do not need maintenance, and the results are stored in the memory of the inspection equipment.

[0044] If a certain subway track area that needs to be inspected is marked as a subway track area that needs maintenance, the geographical location information, topographic image information and warning defect type of the area are further obtained. The above information is compressed and bundled to generate a subway track maintenance information data packet, which is stored in the memory of the inspection equipment, thereby providing comprehensive and accurate data support for the maintenance of the subway track.

[0045] This method can not only quickly and accurately identify the subway track areas that need maintenance, but also further obtain detailed maintenance information (such as geographical location, topography and defect type), generate maintenance information data packets and store them in the inspection equipment, thereby providing comprehensive and accurate data support for the maintenance of subway tracks, improving the efficiency and accuracy of inspection work, and ensuring the safety and normal operation of subway tracks.

[0046] In this embodiment, when the inspection device moves to the sub-subway track area to be inspected, the subway track characteristic image of the sub-subway track area to be inspected is obtained by the camera mechanism carried by the inspection device, specifically: The linear acceleration and angular velocity data of the inspection equipment in three-dimensional space are collected in real time through the three-axis acceleration sensor and the six-degree-of-freedom inertial measurement unit. The motion data of the inspection equipment is obtained by combining the dynamic parameters of pitch, roll and yaw angles obtained by the gyroscope. A multimodal data fusion algorithm is used to perform Kalman filtering fusion on the motion data and the displacement of the visual feature points of the camera mechanism, and a composite feature matrix including the device motion vector and the ambient vibration spectrum is constructed. The motion trajectory of the preset frame image is dynamically fitted through a sliding window mechanism to identify the non-autonomous motion component signal caused by mechanical vibration. Using frequency domain decomposition technology, the non-autonomous motion component signal is decomposed into three characteristic sub-bands: the device active motion fundamental frequency, the mechanical resonance frequency band and the environmental interference harmonics through fast Fourier transform and combining the composite characteristic matrix; An inter-frame motion compensation model is established based on image optical flow field analysis, an affine transformation matrix of adjacent frames is calculated by feature point matching, the feature subband and the affine transformation matrix are introduced into the inter-frame motion compensation model for parameter compensation, and after compensation, a displacement compensation parameter difference is obtained; If the displacement compensation parameter difference is greater than the preset deviation value threshold, a deep learning-assisted vibration recognition mechanism is introduced to input the vibration spectrum features into the pre-trained convolutional neural network for pattern classification, and dynamically adjust the filter parameter weights; until the displacement compensation parameter difference is less than the preset difference threshold, the compensated image sequence is output; The multi-exposure fusion technology is used to perform spatiotemporal fusion processing on the compensated image sequence, the non-local mean denoising algorithm is used to eliminate residual blur, and redundant frames are automatically removed through adaptive threshold detection to output a subway track feature image sequence with spatial continuity and temporal consistency.

[0047] It should be noted that before taking photos, the linear acceleration and angular velocity data of the inspection equipment in three-dimensional space are collected, and the dynamic parameters of pitch, roll and yaw angles obtained by the gyroscope are combined to obtain the motion data of the inspection equipment. The motion trajectory of the preset frame image is dynamically fitted through the sliding window mechanism to identify the non-autonomous motion component signal caused by mechanical vibration. The sliding window mechanism is used to process dynamically changing data to identify and separate non-autonomous motion signals caused by mechanical vibration. Using frequency domain decomposition technology, the non-autonomous motion component signal is decomposed into three characteristic sub-bands: the fundamental frequency of the active motion of the equipment, the mechanical resonance frequency band, and the harmonics of environmental interference through fast Fourier transform. Through frequency domain analysis, complex vibration signals are decomposed into different frequency components for subsequent processing. An inter-frame motion compensation model is established based on image optical flow field analysis, and the affine transformation matrix of adjacent frames is calculated through feature point matching. The characteristic sub-bands and the affine transformation matrix are imported into the inter-frame motion compensation model for parameter compensation. If the displacement compensation parameter difference is greater than the preset deviation value threshold, a deep learning-assisted vibration recognition mechanism is introduced, and the vibration spectrum features are input into the pre-trained convolutional neural network for pattern classification, and the filter parameter weights are dynamically adjusted. When the compensated image quality does not meet the requirements, the deep learning method is used to further optimize the processing of the vibration signal. The compensated image sequence is subjected to spatiotemporal fusion processing, the non-local mean denoising algorithm is used to eliminate residual blur, and redundant frames are automatically removed through adaptive threshold detection. Through multi-exposure fusion and denoising, a high-quality, spatially continuous and temporally consistent subway track feature image sequence is finally generated. Through this method, the motion data of the inspection equipment in the sub-subway track area to be inspected can be accurately obtained, and the non-autonomous motion caused by mechanical vibration and the like can be analyzed and compensated, and finally a subway track feature image sequence with high definition, spatial continuity and temporal consistency is output, which effectively improves the accuracy and reliability of the acquisition of subway track feature images, reduces the influence of interference factors, provides high-quality image data support for the inspection and analysis of subway tracks, and improves the reliability of the inspection results obtained by the inspection equipment.

[0048] In this embodiment, the inspection equipment control method further includes the following steps: The multi-source sensors collect the flight attitude data, environmental parameters and dynamic weight change values ​​of the inspection equipment in real time, including three-dimensional acceleration, angular velocity, GPS location information, barometer altitude data, wind speed and direction sensor readings, air density sensor data and remaining power information provided by the battery management system; The flight attitude data, environmental parameters and dynamic change values ​​of equipment weight are integrated and processed, the GPS position error is corrected using the laser radar point cloud data, and the current three-dimensional spatial coordinates of the inspection equipment are calculated in combination with the posture data of the inertial navigation system; Directly incorporate wind speed and direction sensor readings and air density sensor data into the environmental state matrix, and import the remaining power information provided by the battery management system as key parameters into the environmental state matrix, thereby constructing a multi-dimensional environmental state matrix that includes the current three-dimensional spatial coordinates of the device, wind speed and direction, air density, and remaining battery power; A nonlinear energy consumption function model with lift coefficient, drag coefficient and motor efficiency as variables is established in advance, and the power loss curves at different flight speeds are calculated in real time through the nonlinear energy consumption function model to obtain the theoretical minimum energy consumption cruising speed range; An adaptive weight allocation mechanism is introduced to dynamically adjust the priority weights of the three control variables, namely, flight altitude, heading angle deflection rate and hovering time, according to the theoretical minimum energy consumption cruising speed range and the wind speed fluctuation, orbit geometry complexity and remaining power threshold in the multi-dimensional environmental state matrix, so as to generate a flight parameter combination matching the real-time working conditions; Using genetic algorithms, with the global minimum energy consumption as the objective function, combined with the kinematic constraints of the inspection equipment, obstacle avoidance safety distance restrictions, and flight parameter combinations that match the real-time working conditions, the preset inspection path is dynamically replanned in sections, and the optimal acceleration vector and pitch angle increment of each waypoint are iteratively calculated to generate a smooth flight posture with optimal energy. The inspection equipment is controlled to perform smooth flight according to the energy-optimal smooth flight attitude, ensuring that it always operates within the theoretical minimum energy consumption range during the entire inspection process, thereby achieving the lowest flight energy consumption.

[0049] In summary, this method ensures that the inspection equipment is always in the optimal energy consumption state during the entire inspection process by dynamically adjusting the flight parameters and attitude, thereby improving the energy consumption efficiency during the inspection process, reducing energy consumption, extending the service life of the equipment, and improving the overall efficiency of the inspection work, maximizing the single inspection distance of the inspection equipment, and improving resource utilization.

[0050] The above are only specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A method for controlling intelligent inspection equipment for subway track line inspection, characterized in that: The following steps are involved: Obtaining the degradation index of the subway track to be inspected under the combined effect of various environmental factor data, and constructing a degradation index hash dictionary of the subway track to be inspected according to the degradation index of the subway track to be inspected under the combined effect of various environmental factor data; The subway track to be inspected is divided into several sub-subway track areas, and a historical dynamic environmental factor data set of each sub-subway track area within a preset time period is obtained. According to the historical dynamic environmental factor data set and in combination with a degradation index hash dictionary, a degradation index evolution curve of each sub-subway track area attenuated and aggregated over time is constructed; Each sub-subway track area is evaluated according to the degradation index evolution curve of each sub-subway track area attenuated over time, and the dynamic risk score value of each sub-subway track area is obtained; the sub-subway track area that needs to be inspected and the sub-subway track area that does not need to be inspected are determined according to the dynamic risk score value of each sub-subway track area; Obtaining geographical location information of each subway track area that needs to be inspected, and generating an inspection route for the inspection equipment according to the geographical location information of each subway track area that needs to be inspected; Controlling the inspection equipment to inspect the subway track to be inspected according to the inspection route; When the inspection device moves to the sub-subway track area to be inspected, a subway track feature image of the sub-subway track area to be inspected is obtained through a camera mechanism carried by the inspection device, and features are extracted from the subway track feature image to obtain an LBP feature matrix of the sub-subway track area to be inspected; The sub-subway track area to be inspected is analyzed according to the LBP feature matrix of the sub-subway track area to be inspected, an inspection result is generated, and the inspection result is transmitted to the memory of the inspection device for storage.

2. The intelligent inspection equipment control method for subway track line inspection according to claim 1 is characterized in that: Obtain the degradation index of the subway track to be inspected under the combined effect of various environmental factor data, and construct a degradation index hash dictionary of the subway track to be inspected according to the degradation index of the subway track to be inspected under the combined effect of various environmental factor data, specifically: Acquire rail material information of the subway track to be inspected, and retrieve the degradation index of the subway track to be inspected under the combined effect of various environmental factor data in the big data network according to the rail material information; And build a dynamic hash map; Discretize each environmental factor data combination to obtain the factor discrete value of each environmental factor data combination; and combine each factor discrete value into a binary feature code according to a preset weight to generate a unique hash identification key for each environmental factor data combination; Generate a number of key-value pairs in the dynamic hash mapping table according to the hash identification key corresponding to each environmental factor data combination; associate the degradation index under the action of each environmental factor data combination with the corresponding hash identification key to perform key-value pair mapping to obtain the linked list storage slot mapped by each hash identification key; The corresponding degradation indexes are written into the linked list storage slots mapped by the corresponding hash identification keys in the dynamic hash mapping table through the double hash conflict resolution mechanism, and the degradation index hash dictionary of the subway track to be inspected is obtained.

3. The intelligent inspection equipment control method for subway track line inspection according to claim 1 is characterized in that: The subway track to be inspected is divided into several sub-subway track areas, and a historical dynamic environmental factor dataset of each sub-subway track area within a preset time period is obtained. According to the historical dynamic environmental factor dataset and in combination with the degradation index hash dictionary, a degradation index evolution curve of each sub-subway track area attenuated and aggregated over time is constructed, specifically: Perform time series feature analysis on the historical dynamic environmental factor data set of each sub-subway track area within a preset time period, divide the dynamic timestamps based on the sliding window mechanism, extract the corresponding historical dynamic environmental factor data at each timestamp, and obtain the historical dynamic environmental factor data of each sub-subway track area at each timestamp; Convert the historical dynamic environmental factor data of each sub-subway track area at each time stamp into a corresponding binary feature code to obtain a hash identification key corresponding to the historical dynamic environmental factor data of each sub-subway track area at each time stamp; Importing the hash identification key corresponding to the historical dynamic environmental factor data of each sub-subway track area at each time stamp into the degradation index hash dictionary; In order of timestamps, the similarities between the hash identification key corresponding to the historical dynamic environmental factor data of the corresponding timestamp and the hash identification key stored in each key-value pair in the degradation index hash dictionary are calculated in sequence, and the key-value pair with the highest similarity is marked, and the degradation index of the sub-subway track area at the corresponding timestamp is extracted from the linked list storage slot of the marked key-value pair; By analogy, the degradation index of each sub-subway track area based on the timestamp order is obtained; An exponential decay function is preset, and based on the exponential decay function, the degradation index of each sub-subway track area based on the timestamp order is time-weighted, and an evolution curve of the degradation index of each sub-subway track area that decays and aggregates over time is output.

4. The intelligent inspection equipment control method for subway track line inspection according to claim 1 is characterized in that: Each sub-subway track area is evaluated according to the degradation index evolution curve of each sub-subway track area over time, and the dynamic risk score value of each sub-subway track area is obtained, which is as follows: Perform time window sliding segmentation on the degradation index evolution curve of each sub-subway track area to extract local time series segments; calculate the ratio of the standard deviation to the mean of the local time series segments to obtain the coefficient of variation of each sub-subway track area; and calculate the first-order derivative and the second-order derivative of the degradation index evolution curve of each sub-subway track area; The coefficient of variation, first-order derivative and second-order derivative of the degradation curve of each sub-region are synchronized by time stamp through the multi-source data spatiotemporal alignment algorithm to obtain the synchronized coefficient of variation, synchronized first-order derivative and synchronized second-order derivative under the same time reference; Deploy a multi-dimensional feature fusion channel and use spatiotemporal convolution kernels to extract the fluctuation frequency domain characteristics of the synchronized coefficient of variation, the trend directional characteristics of the synchronized first-order derivative, and the degradation acceleration characteristics of the synchronized second-order derivative. A multi-head attention mechanism is introduced to perform cross-dimensional interaction on the fluctuation frequency domain characteristics of the synchronized coefficient of variation, the trend directional characteristics of the synchronized first-order derivative, and the degradation acceleration characteristics of the synchronized second-order derivative, generating a fused feature tensor containing spatiotemporal correlation weights. The fused feature tensor is input into the deep feature fusion network, and the multi-scale spatiotemporal correlation pattern is extracted using residual connection and feature pyramid structure, and a three-dimensional risk indicator vector including volatility intensity, trend stability and acceleration probability is output; The three-dimensional risk index vector is mapped to a preset risk level space, and the discrete levels are nonlinearly weighted aggregated through a radial basis function neural network to generate a dynamic risk score value for each sub-subway track area.

5. The intelligent inspection equipment control method for subway track line inspection according to claim 1 is characterized in that: According to the dynamic risk score of each sub-subway track area, the sub-subway track area that needs to be inspected and the sub-subway track area that does not need to be inspected are determined as follows: Obtaining a dynamic risk score value of each sub-subway track area and setting a dynamic risk score value threshold; determining whether the dynamic risk score value of each sub-subway track area is greater than the dynamic risk score value threshold; If the dynamic risk score value of a certain sub-subway track area is greater than the dynamic risk score value threshold, the sub-subway track area is determined as a sub-subway track area that needs to be inspected; If the dynamic risk score value of a certain sub-subway track area is not greater than the dynamic risk score value threshold, the sub-subway track area is determined as a sub-subway track area that does not require inspection.

6. The intelligent inspection equipment control method for subway track line inspection according to claim 1 is characterized in that: Obtain the geographical location information of each subway track area that needs to be inspected, and generate the inspection route of the inspection equipment according to the geographical location information of each subway track area that needs to be inspected, specifically: Obtain a track distribution map of the subway track to be inspected, and mark the geographical location information of each sub-subway track area to be inspected in the track distribution map; The particle swarm optimization algorithm is introduced to construct the initial solution space based on the geographical location information of each subway track area to be inspected. Each sub-area is abstracted as a particle position in the particle swarm optimization algorithm, and the speed and position distribution of the particle swarm are initialized according to the preset moving speed and starting position of the inspection equipment. By continuously distributing the speed and position of the current particle group, each particle is moved and updated in the solution space according to the preset rules. After a preset number of iterations, several different moving paths of particles from the starting position to each sub-subway track area to be inspected are obtained; Calculate the total path length value of each moving path; The moving path with the shortest total path length value is used as the inspection route of the inspection equipment.

7. The intelligent inspection equipment control method for subway track line inspection according to claim 1 is characterized in that: Feature extraction is performed on the subway track feature image to obtain the LBP feature matrix of the sub-subway track area to be inspected, specifically: The collected subway track images are preprocessed by grayscale conversion and illumination balance to enhance local texture contrast; a dynamic radius strategy is used to generate annular sampling points within the pixel neighborhood; A multi-scale differential comparison mechanism is introduced to perform secondary threshold quantization on the grayscale difference between the central pixel and the ring sampling point to generate a rotationally invariant binary code sequence. The local texture descriptor is constructed through a sliding window, and the direction is weighted and fused in combination with the track direction to eliminate the feature offset caused by mechanical vibration. The whole image is scanned in blocks and overlapped, and the multi-level LBP response images are integrated and normalized dimension reduction is performed to generate an LBP feature matrix with spatial distribution characteristics.

8. The intelligent inspection equipment control method for subway track line inspection according to claim 1 is characterized in that: The sub-subway track area to be inspected is analyzed according to the LBP feature matrix of the sub-subway track area to be inspected, an inspection result is generated, and the inspection result is transmitted to the memory of the inspection device for storage, specifically: Obtaining LBP feature matrices corresponding to various early warning defects of subway tracks through a big data network, building a database, and importing the LBP feature matrices corresponding to various early warning defects of subway tracks into the database; After obtaining the LBP feature matrix of a certain subway track area that needs to be inspected, importing the LBP feature matrix of the subway track area that needs to be inspected into the database; Calculate the matrix overlap between the LBP feature matrix of the sub-subway track area to be inspected and each LBP feature matrix in the database; If the matrix overlap between the LBP feature matrix of the sub-subway track area to be inspected and each LBP feature matrix in the database is not greater than the preset overlap threshold, it means that there is no warning defect in the sub-subway track area to be inspected, and the sub-subway track area to be inspected is marked as a subway track area that does not require maintenance; If the matrix overlap between the LBP feature matrix of the sub-subway track area to be inspected and one or more LBP feature matrices in the database is greater than a preset overlap threshold, it means that there is a warning defect in the sub-subway track area to be inspected, and the sub-subway track area to be inspected is marked as a subway track area that needs maintenance; An inspection result is generated according to the subway track area that needs maintenance and the subway track area that does not need maintenance, and the inspection result is transmitted to a memory of the inspection device for storage.

9. The intelligent inspection equipment control method for subway track line inspection according to claim 8, characterized in that: The following steps are also included: If a certain sub-subway track area requiring inspection is marked as a subway track area requiring maintenance, obtaining geographic location information of the subway track area requiring maintenance; At the same time, the inspection equipment is controlled to obtain the topographic image information of the subway track area that needs maintenance within a preset radius; and the warning defect type of the subway track area that needs maintenance; The geographical location information, topographic image information and warning defect type of the subway track area requiring maintenance are compressed and bundled to generate a subway track maintenance information data packet, and the subway track maintenance information data packet is transmitted to the memory of the inspection equipment for storage.

Images