Railway environment identification method and system based on artificial intelligence

Abstract

The invention provides a railway line environment identification method and system based on artificial intelligence. According to the method, the laser point cloud, the track inspection vehicle state parameters and the millimeter wave radar data of the railway ballast area are synchronously collected, and the enhanced point cloud with the physical attribute is fused and constructed. Linear scanning laser and millimeter wave radar at the bottom of a rail inspection vehicle are used for spiral scanning and penetrating scanning, and a three-dimensional data matrix fusing the surface appearance and the internal dielectric characteristics is generated. The neural network analyzes the railway ballast contact surface curvature and gap distribution of the matrix, and enhances the weight of the loose area through space-time attention. And a model is generated based on adversarial training to complement the occlusion area, and a complete virtual point cloud is output. And finally, quantitatively analyzing the virtual point cloud and related characteristics, and outputting the railway ballast compactness and the coordinate range of the hardened area as key identification results. Through multi-source data fusion, deep learning and point cloud generation technologies, accurate, dynamic and non-contact quantitative evaluation of the railway ballast compactness and the hardened area is realized.

Description

Technical Field

[0001] The present application relates to the field of railway intelligent detection technology, and in particular to a method and system for identifying railway environment based on artificial intelligence. Background Art

[0002] The density of railway ballast, or the crushed stone layer beneath the railroad sleepers, and whether it is loose or compacted, with the gravel hardened by mud, are key factors directly affecting track geometric stability and driving safety. Traditional manual inspections are inefficient, subjective, incomplete, and pose safety risks. An automated, non-contact, highly accurate, and surface-penetrating intelligent detection technology is urgently needed to rapidly and quantitatively identify and locate ballast density and compacted areas along railway lines, replacing manual inspections and improving operational efficiency and safety.

[0003] Currently, a representative technical solution involves point cloud acquisition and machine learning recognition based on vehicle-mounted 3D laser scanning. This solution utilizes a laser scanner mounted on a track inspection vehicle to quickly acquire high-density point cloud data of the track and ballast surfaces. Ballast regions are extracted using point cloud processing algorithms and further analyzed using machine learning models, surface roughness, or geometric features to differentiate between "normal," "loose," and "compacted" areas.

[0004] This solution mainly relies on surface morphology information and has significant limitations: First, it cannot sense the internal structure of the ballast: lasers have difficulty penetrating gravel, and cannot obtain key internal physical properties such as the distribution of gaps between ballast particles, changes in internal density, and internal bonding of the compacted material. The judgment of the "loose" state is mainly based on the surface concave and convex appearance, which is easily disturbed by surface dirt and water accumulation, and the accuracy of early or deep compaction / looseness identification is insufficient; second, the occlusion problem is serious: surface gravel will block the underlying information, resulting in a large amount of missing point cloud data. Existing methods cannot effectively reconstruct the true accumulation state of ballast in the obscured area; third, the quantification ability is weak: usually only rough classification results can be given, and there is a lack of accurate numerical output of ballast density and precise definition of the spatial coordinate range of the boundary and depth of the compacted area, which makes it difficult to meet the needs of refined maintenance decision-making. Summary of the Invention

[0005] The present application provides an artificial intelligence-based railway environment recognition method and system to solve the problems in the existing technology of relying on surface morphology data and being unable to penetrate and perceive the internal structure of ballast, missing occluded areas and weak quantification ability.

[0006] In a first aspect, the present application provides a method for identifying railway environment based on artificial intelligence, comprising:

[0007] Acquire laser point cloud data of the track ballast area, and simultaneously collect the operating status parameters of the track inspection vehicle and the echo data of the millimeter-wave radar;

[0008] Performing dynamic distortion compensation on the laser point cloud data based on the operating state parameters, and constructing an enhanced point cloud model including physical properties in combination with dielectric constant characteristics in the echo data;

[0009] The surface of the ballast area is spirally scanned by a line-scan laser module at the bottom of the track inspection vehicle, and the millimeter-wave radar is synchronously triggered to perform a penetrating scan to generate a three-dimensional spatial data matrix that integrates the surface topography and internal dielectric characteristics;

[0010] A neural network structure is used to analyze the curvature variation law data of the ballast particle contact surface in the three-dimensional spatial data matrix and identify the spatial distribution characteristics of the gaps between the ballast particles. At the same time, the weight of the loose area is dynamically enhanced through a spatiotemporal attention mechanism.

[0011] A multi-scale point cloud generation model is constructed based on an adversarial training mechanism to geometrically complete the missing occluded areas in the spatial distribution features to output a virtual point cloud of the complete ballast accumulation state;

[0012] The virtual point cloud is quantitatively analyzed, and combined with the dielectric constant distribution characteristics, the curvature change law data and the loose area weight in the enhanced point cloud model, the ballast density value and the spatial coordinate range of the compacted area are output, which serve as the key identification results of the environmental status of the ballast structure along the railway.

[0013] Optionally, the multi-scale point cloud generation model is constructed based on the adversarial training mechanism, and the missing occluded areas in the spatial distribution features are geometrically completed to output a virtual point cloud of a complete ballast accumulation state, including:

[0014] Based on the adversarial training mechanism, a multi-scale point cloud generation model is constructed, which includes a generator unit and a discriminator unit. The generator unit is provided with a coarse-grained generation channel and a fine-grained generation channel, and the discriminator unit is provided with a geometric continuity verification module.

[0015] Inputting the occlusion area in the spatial distribution feature into the generator unit, predicting the basic geometry of the occlusion area through the coarse-grained generation channel, and adding the surface details of the ballast particles through the fine-grained generation channel to generate supplementary point cloud data of the occlusion area;

[0016] The supplementary point cloud data and the original spatial distribution features are input into the discriminator unit, and the curvature connection state of the supplementary geometric body and the adjacent area is detected by the geometric continuity verification module, while the physical rationality of the particle gap transition is verified, and the discrimination result is output;

[0017] When the discrimination result fails to pass the verification, the generator unit parameters are iteratively optimized to regenerate supplementary point cloud data. When the discrimination result passes the verification, the supplementary point cloud data is spliced ​​with the original spatial distribution characteristics to form a three-dimensional point set data of the complete ballast accumulation state as a virtual point cloud output.

[0018] Optionally, dynamically enhancing the weights of loose regions through a spatiotemporal attention mechanism includes:

[0019] A spatiotemporal attention mechanism is constructed based on the spatial distribution characteristics of the gaps between ballast particles;

[0020] Add weight adjustment increments to the gap areas between ballast particles with high frequency spatial position changes, and apply time accumulation enhancement factors to the gap areas between ballast particles that continue to expand over time;

[0021] The weight value of the loose area in the spatial distribution feature is increased based on the weight adjustment increment and the time accumulation enhancement factor.

[0022] Optionally, the quantitative analysis of the virtual point cloud is combined with the dielectric constant distribution characteristics, the curvature change law data and the loose area weight in the enhanced point cloud model to output the ballast density value and the spatial coordinate range of the compacted area, which are used as key identification results of the environmental status of the ballast structure along the railway, including:

[0023] Performing quantitative analysis on the virtual point cloud, calculating the point density distribution of the virtual point cloud within a unit volume spatial grid, and using the median value of the point density distribution as a ballast density reference value;

[0024] Combined with the dielectric constant distribution characteristics, curvature change law data and loose area weight value in the enhanced point cloud model, a hardened area determination condition is established. The hardened area determination condition includes three independent determination conditions: the first determination condition is to mark the area in the enhanced point cloud model where the dielectric constant characteristic value is continuously higher than the set threshold; the second determination condition is to extract the flat area where the curvature change amount is lower than the curvature threshold in the curvature change law data; the third determination condition is to locate the area with a weight value lower than the critical weight value in the loose area weight value distribution;

[0025] Determine an area that satisfies the first, second, and third determination conditions simultaneously as a hardened area, extract the coordinates of spatial turning points of the outer contour of the hardened area, and connect the coordinates of the spatial turning points to form a three-dimensional polygonal bounding box;

[0026] The ballast density reference value is output as the ballast density value, and the vertex coordinate set of the three-dimensional polygonal bounding box is output as the spatial coordinate range of the compacted area. The ballast density value and the spatial coordinate range of the compacted area together constitute the key identification result of the environmental status of the ballast structure along the railway.

[0027] Optionally, the line scanning laser module at the bottom of the track inspection vehicle performs a spiral scan on the surface of the ballast area, and synchronously triggers the millimeter wave radar to perform a penetrating scan to generate a three-dimensional spatial data matrix that integrates the surface topography and internal dielectric characteristics, including:

[0028] Based on the forward direction control, the galvanometer deflection assembly of the line scanning laser module at the bottom of the track inspection vehicle moves along the Archimedean spiral trajectory, driving the laser beam to cover the surface of the ballast area in a spiral path, and generating a three-dimensional coordinate point set of the ballast surface topography by receiving the reflected laser beam;

[0029] At the start of the laser beam scan, a trigger pulse is sent to the millimeter-wave radar controller to drive the millimeter-wave radar to emit a penetrating beam. After receiving the reflected signal from the ballast layer, the dielectric constant characteristic values ​​at different depths of the ballast layer are extracted.

[0030] Based on the surface three-dimensional coordinate point set, the surface coordinate point is vertically projected onto the underground coordinate system, and the dielectric constant characteristic value of each depth layer below the projection point is associated with the projection position to establish a spatial mapping relationship between the surface three-dimensional coordinate point and the underground depth layer;

[0031] Based on the spatial mapping relationship, the data is integrated into a spatial grid, the surface height value and the corresponding underground dielectric constant profile value are stored with the surface projection point as the center position, and all spatial grid units are aggregated to generate a three-dimensional spatial data matrix.

[0032] Optionally, the adopting of a neural network structure to analyze the curvature variation regularity data of the contact surface of the ballast particles in the three-dimensional spatial data matrix and identifying the spatial distribution characteristics of the gaps between the ballast particles includes:

[0033] Processing the three-dimensional spatial data matrix through a neural network structure, extracting a surface normal vector change sequence of the ballast particle contact surface area in a primary processing module of the neural network structure, calculating a curvature change quantization value based on the surface normal vector change sequence, and generating data on the curvature change pattern of the ballast particle contact surface;

[0034] In the advanced processing module of the neural network structure, a continuous spatial region with a constant curvature quantization value in the three-dimensional spatial data matrix is ​​detected, the continuous spatial region is identified as a set of ballast particle gaps, the three-dimensional spatial distribution parameters of the set of ballast particle gaps are measured, and the spatial distribution characteristics of the ballast particle gaps are generated.

[0035] Optionally, the performing dynamic distortion compensation on the laser point cloud data based on the operating state parameters and constructing an enhanced point cloud model including physical properties in combination with dielectric constant characteristics in the echo data includes:

[0036] Generate a position compensation vector for each three-dimensional coordinate point in the laser point cloud data based on the longitudinal displacement, lateral offset and pitch angle values ​​in the operating state parameters;

[0037] According to the position compensation vector, the original coordinates of the laser point cloud data are subjected to point-by-point translation correction to obtain a corrected three-dimensional coordinate point set;

[0038] Analyzing the dielectric constant characteristic values ​​at different depths of the ballast layer from the millimeter-wave radar echo data, and mapping the dielectric constant characteristic values ​​as new physical properties to corresponding points in the corrected three-dimensional coordinate point set according to spatial positions;

[0039] The attribute dimension is expanded for each corrected 3D coordinate point to generate a composite data unit containing both spatial coordinates and dielectric constant values. All composite data units are aggregated to form an enhanced point cloud model.

[0040] In a second aspect, the present application provides an artificial intelligence-based railway environment recognition system, comprising:

[0041] The acquisition module is used to obtain laser point cloud data of the track ballast area and synchronously collect the operating status parameters of the track inspection vehicle and the echo data of the millimeter wave radar;

[0042] A construction module, configured to perform dynamic distortion compensation on the laser point cloud data based on the operating state parameters, and to construct an enhanced point cloud model including physical properties in combination with dielectric constant characteristics in the echo data;

[0043] A generation module is used to perform a spiral scan on the surface of the ballast area using a line scanning laser module at the bottom of the track inspection vehicle, and synchronously trigger the millimeter wave radar to perform a penetration scan to generate a three-dimensional spatial data matrix that integrates the surface topography and internal dielectric characteristics;

[0044] an analysis module for analyzing the curvature variation law data of the contact surface of the ballast particles in the three-dimensional spatial data matrix using a neural network structure and identifying the spatial distribution characteristics of the gaps between the ballast particles, while dynamically enhancing the weight of loose areas through a spatiotemporal attention mechanism;

[0045] A completion module is used to build a multi-scale point cloud generation model based on an adversarial training mechanism, and geometrically complete the missing occluded areas in the spatial distribution features to output a virtual point cloud of the complete ballast accumulation state;

[0046] The output module is used to perform quantitative analysis on the virtual point cloud, combine the dielectric constant distribution characteristics in the enhanced point cloud model, the curvature change law data and the loose area weight, output the ballast density value and the spatial coordinate range of the compacted area, and use it as the key identification result of the environmental status of the ballast structure along the railway.

[0047] In a third aspect, the present application provides a computing device comprising a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are used to be called and executed by the processing component to implement an artificial intelligence-based railway environment identification method as described in the first aspect above.

[0048] In a fourth aspect, the present application provides a computer storage medium storing a computer program. When the computer program is executed by a computer, it implements a method for identifying railway environment based on artificial intelligence as described in the first aspect.

[0049] In the example of this application, the laser point cloud data of the track ballast area is obtained, and the operating status parameters of the track inspection vehicle and the echo data of the millimeter wave radar are collected synchronously; the laser point cloud data is dynamically distorted based on the operating status parameters, and the dielectric constant characteristics in the echo data are combined to construct an enhanced point cloud model containing physical properties; the surface of the ballast area is spirally scanned by the line scanning laser module at the bottom of the track inspection vehicle, and the millimeter wave radar is synchronously triggered to perform a penetration scan to generate a three-dimensional spatial data matrix that integrates the surface morphology and internal dielectric characteristics; a neural network structure is used to analyze the ballast in the three-dimensional spatial data matrix. The curvature change law data of the particle contact surface is used to identify the spatial distribution characteristics of the gaps between ballast particles, and the weight of loose areas is dynamically enhanced through the spatiotemporal attention mechanism. A multi-scale point cloud generation model is constructed based on the adversarial training mechanism, and the missing occluded areas in the spatial distribution characteristics are geometrically completed to output a virtual point cloud of the complete ballast accumulation state. The virtual point cloud is quantitatively analyzed, and the dielectric constant distribution characteristics, the curvature change law data and the loose area weights in the enhanced point cloud model are combined to output the ballast density value and the spatial coordinate range of the compacted area, which are used as the key identification results of the ballast structure environmental status along the railway.

[0050] The technical solution of this application has the following beneficial effects:

[0051] This application synchronously collects laser point clouds, track inspection vehicle status parameters and millimeter-wave radar echo data, first compensates for the dynamic distortion of the point cloud based on the status parameters, and integrates the radar dielectric constant to construct a point cloud model with enhanced physical properties; then, through the synchronous triggering of line-scan laser spiral scanning and millimeter-wave radar penetration scanning, a three-dimensional spatial data matrix that integrates the surface morphology and internal dielectric characteristics is generated; then, a neural network is used to analyze the curvature changes and gap distribution characteristics of the ballast particle contact surface, and the spatiotemporal attention mechanism is combined to dynamically enhance the weight of loose areas; a multi-scale point cloud generation model is constructed based on adversarial training, and the occluded areas are geometrically completed to output a virtual point cloud of the complete ballast accumulation; finally, through quantitative analysis of the virtual point cloud, combined with the dielectric constant distribution, curvature change law and loose area weight, the ballast density value and the spatial coordinate range of the compacted area are accurately output, realizing non-contact, penetrating, and high-precision quantitative identification of the internal structural state of railway ballast, forming an automated intelligent detection closed loop to replace manual inspections.

[0052] A multi-scale point cloud generation model is further constructed, consisting of a dual-channel generator with coarse-grained and fine-grained generation channels and a geometric continuity verification discriminator. The occluded area is input into the generator, where the basic geometry is first predicted through the coarse-grained channel, and then the surface details of the ballast particles are added through the fine-grained channel to generate a supplementary point cloud. The supplementary point cloud and the original data are input into the discriminator, and the geometric continuity verification module is used to check the connection between the supplementary area and the adjacent curvature and the physical rationality of the particle gaps. If the verification fails, the generator parameters are iteratively optimized and regenerated. After passing the verification, the supplementary point cloud is spliced ​​with the original data to output a complete 3D virtual point cloud. This method achieves hierarchical completion of "basic geometric framework construction-granular detail restoration" through the dual-scale generation channel. Combined with the dual constraints of curvature connection and gap physical rationality imposed by the geometric continuity verification module, it ensures that the supplemented point cloud in the occluded area conforms to the macroscopic topological structure of the ballast accumulation while maintaining the physical authenticity of the particle-level microscopic features. This results in a complete ballast virtual point cloud with high geometric consistency and physical rationality, providing a reliable data foundation for subsequent quantitative analysis.

[0053] These and other aspects of the present application will become more readily apparent from the description of the following embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, a brief introduction will be given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0055] Figure 1A flowchart of a method for identifying railway environment based on artificial intelligence provided by the present application is shown;

[0056] Figure 2 A scene diagram showing a method for identifying railway environment based on artificial intelligence provided by the present application is shown;

[0057] Figure 3 The present invention provides a schematic diagram of the structure of an artificial intelligence-based railway environment recognition system;

[0058] Figure 4 A schematic structural diagram of a computing device provided by the present application is shown. DETAILED DESCRIPTION

[0059] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application.

[0060] In some of the processes described in the specification and claims of this application and the above-mentioned figures, multiple operations that appear in a specific order are included, but it should be clearly understood that these operations may not be executed in the order in which they appear in this document or may be executed in parallel. The serial numbers of the operations, such as 101, 102, etc., are only used to distinguish between different operations, and the serial numbers themselves do not represent any order of execution. In addition, these processes may include more or fewer operations, and these operations may be executed in sequence or in parallel. It should be noted that the descriptions of "first", "second", etc. in this document are used to distinguish different messages, devices, modules, etc., and do not represent a sequential order, nor do they limit "first" and "second" to being different types.

[0061] Research has shown that existing solutions based on vehicle-mounted 3D laser scanning and machine learning for automated railway ballast condition monitoring face fundamental bottlenecks: They rely on single-source surface topography, making it impossible to penetrate and understand key internal physical properties such as the distribution of ballast particle gaps, internal density variations, and the bonding state of the aggregated material. Furthermore, severe surface occlusion results in large areas of missing point cloud data, making it difficult to reconstruct the true ballast accumulation structure. Ultimately, the output is limited to a rough qualitative classification, lacking precise numerical assessments of density and spatial coordinates defining the boundaries and depth of aggregated areas, making it difficult to meet the needs of refined and quantitative maintenance decision-making. The core of this series of shortcomings lies in the triple dilemma of existing technologies: the inability to see the internal structure of the ballast, the inability to know occluded areas, and the inability to measure state parameters.

[0062] To address the above issues, this application proposes an artificial intelligence-based method for identifying railway environment. Its core lies in the synchronous fusion of laser point clouds, millimeter-wave radar echoes, and track inspection vehicle status parameters to construct an enhanced point cloud model containing physical properties; through the synchronous triggering of line-scan laser spiral scanning and millimeter-wave penetration scanning, a three-dimensional spatial data matrix is ​​generated; then, a neural network is used to analyze the particle contact curvature and gap distribution characteristics, and the spatiotemporal attention mechanism is used to dynamically focus on loose areas; in particular, an adversarial training mechanism is applied to construct a multi-scale point cloud generation model, and the occluded areas are intelligently completed with geometric and physical dual constraints, outputting a virtual point cloud of the complete ballast accumulation; finally, the virtual point cloud is quantitatively analyzed, and the dielectric constant, curvature law, and loose weight are combined to accurately output the ballast density value and the spatial coordinate range of the compacted area. This method breaks through the triple dilemma of background technology in one fell swoop. It makes the internal structure "visible" through the fusion of millimeter wave penetration and dielectric constant, makes the occluded area "knowable" through the geometric-physical completion of the adversarial generative model, and outputs "measurable" density values ​​and precise spatial coordinates through the quantitative analysis engine of multi-feature fusion, thus providing a reliable technical closed loop for the automated, precise and quantitative detection of railway ballast status, and solving the problems of insufficient recognition accuracy and lack of quantification ability caused by the existing solutions' reliance on surface information.

[0063] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without making creative efforts are within the scope of protection of this application.

[0064] Figure 1 A flowchart of a method for identifying railway environment based on artificial intelligence is provided for the embodiment of the present application, such as Figure 1 As shown, the method includes:

[0065] 101. Obtain laser point cloud data of the track ballast area, and simultaneously collect the operating status parameters of the track inspection vehicle and the echo data of the millimeter wave radar;

[0066] In the above solution, laser point cloud data refers to a dense set of three-dimensional coordinate points generated by transmitting a laser beam through the line-scan laser module at the bottom of the track inspection vehicle and receiving the reflected signal from the ballast surface. Each point contains spatial position (X, Y, Z) information and is used to characterize the geometric shape of the ballast surface. Operational status parameters refer to the dynamic physical quantities of the track inspection vehicle during movement, including vehicle speed, longitudinal acceleration, lateral acceleration, pitch angle, and roll angle. They are used to subsequently correct data distortion caused by vehicle shaking during laser scanning. Millimeter-wave radar echo data refers to the sequence generated by the vehicle-mounted millimeter-wave radar transmitting electromagnetic waves to the ballast and receiving the reflected signal. It contains signal strength, phase difference, and propagation time. The dielectric constant inside the ballast can be extracted through analysis.

[0067] In the embodiment of the present application, first, a laser beam is emitted at a frequency of 200 Hz by the line scanning laser module at the bottom of the track inspection vehicle to perform high-speed line scanning on the ballast surface on both sides of the track. When the laser beam hits the ballast particles, the receiver records the time difference and angle of the reflected light, and combines the spatial positioning data of the built-in high-precision gyroscope to generate a single frame of surface point cloud containing 50,000 to 80,000 three-dimensional coordinate points as laser point cloud data. For example, the laser beam scans a raised gravel on the ballast surface, calculates the distance between the point and the sensor based on the round-trip time difference of the light beam, and determines its precise position (X=1.2m, Y=0.3m, Z=0.15m) in combination with the scanning angle to form a data point in the point cloud.

[0068] Secondly, the vehicle's operating parameters are collected at a 100Hz frequency using the onboard IMU (Inertial Measurement Unit). These parameters include vehicle speed, triaxial acceleration, pitch angle, and roll angle. These parameters are packaged into a data packet with a synchronized timestamp. For example, when the vehicle bumps past a track joint, the IMU records a sudden change in longitudinal acceleration to +0.08g and a pitch angle change of +0.8°. This data packet is timestamped at T = 12.345s.

[0069] Finally, the millimeter-wave radar is triggered to emit a 77GHz frequency-modulated continuous wave. After penetrating the ballast surface, the electromagnetic wave is reflected by various media. The receiver captures the echo signal, analyzes its amplitude attenuation and phase shift, and outputs a sequence of raw echoes containing a time-intensity relationship. For example, when the radar wave penetrates the ballast, the echo amplitude between dry gravel is attenuated by 20dB, while the amplitude attenuation in water-filled, compacted areas is only 5dB due to the increased dielectric constant, forming a characteristic signal that can distinguish the internal structure.

[0070] In a practical application, during routine inspections on a freight railway, track inspection vehicle A travels at 25 km / h to section B. The undercarriage laser module scans the ballast area, generating a surface point cloud with 72,000 points per frame. Simultaneously, the IMU records the vehicle's speed of 25 km / h, lateral acceleration of 0.05 g, and pitch angle of -0.6°. Simultaneously, the millimeter-wave radar emits electromagnetic waves, collecting echo signals lasting 5 milliseconds. It detects an abnormally high-amplitude echo in section C, at coordinates X = 120m to 125m, indicating a risk of compaction. All three data types are timestamped with a unified "2023-05-10 14:30:25.123" for precise synchronization.

[0071] The 101 overall solution, through simultaneous multi-source sensing, acquires data on the ballast surface's fine geometry, vehicle dynamics, and internal physical properties in a single inspection, laying the foundation for subsequent fusion processing. Its core value lies in overcoming the limitations of a single sensor. Lasers accurately depict the surface but cannot penetrate the interior, while radars can perceive the interior but lack geometric precision. State parameters provide the basis for dynamic distortion compensation. These three elements work together to provide comprehensive data support covering "surface, interior, and motion."

[0072] 102. Perform dynamic distortion compensation on the laser point cloud data based on the operating state parameters, and construct an enhanced point cloud model including physical properties in combination with dielectric constant characteristics in the echo data;

[0073] Optionally, step 102 may specifically include the following steps:

[0074] 1021. Generate a position compensation vector for each three-dimensional coordinate point in the laser point cloud data based on the longitudinal displacement, lateral offset, and pitch angle value in the operating state parameters;

[0075] 1022. Perform point-by-point translation correction on the original coordinates of the laser point cloud data according to the position compensation vector to obtain a corrected three-dimensional coordinate point set;

[0076] 1023. Analyze the dielectric constant characteristic values ​​at different depths of the ballast layer from the millimeter-wave radar echo data, and map the dielectric constant characteristic values ​​as new physical attributes to corresponding points in the corrected three-dimensional coordinate point set according to spatial positions;

[0077] 1024. Expand the attribute dimension for each corrected three-dimensional coordinate point, generate a composite data unit containing both the spatial coordinates and the dielectric constant value, and aggregate all the composite data units to form an enhanced point cloud model.

[0078] In the above scheme, the position compensation vector refers to the three-dimensional spatial offset (Δx, Δy, Δz) calculated based on the motion deviation of the track inspection vehicle, namely the longitudinal displacement, lateral offset, and pitch angle value, which is used to correct the position error of each point in the laser point cloud. The corrected three-dimensional coordinate point set refers to the new point cloud generated after the original laser point cloud is corrected by the position compensation vector, eliminating the deformation caused by vehicle movement. The dielectric constant eigenvalue refers to the physical quantity analyzed from the millimeter-wave radar echo, which represents the conductive properties of the ballast material and is used to distinguish the internal structure. The composite data unit refers to the expanded data structure of each point cloud data point, which contains the dual attributes of spatial coordinates (X, Y, Z) and dielectric constant. The enhanced point cloud model refers to the point cloud set composed of all composite data units, which has both geometric position and internal physical property information.

[0079] In the embodiment of the present application, first, step 1021 is performed based on the operating state parameters of the track inspection vehicle, namely the longitudinal displacement, lateral offset, and pitch angle value, and a kinematic affine transformation model is used to calculate the position compensation vector of each three-dimensional coordinate point in the laser point cloud. The longitudinal displacement is obtained by integrating the vehicle speed. For example, when the vehicle speed is 30km / h and the laser sampling interval is 5ms, the longitudinal compensation Δx=0.042m; the lateral offset is derived by the quadratic integration of the IMU acceleration data, such as a 0.1g lateral acceleration lasting 0.1s produces Δy=0.0049m; the pitch angle compensation is calculated by trigonometric function projection, and a laser installed at a height of 1.5m produces Δz=0.021m at a pitch angle of +0.8°. Finally, the three-dimensional position compensation vector (Δx, Δy, Δz) of each point is output to provide a dynamic offset reference for subsequent correction.

[0080] Secondly, the position compensation vector generated in step 1021 is used to perform point-by-point coordinate correction on the original laser point cloud in step 1022. First, the compensation vector is matched with the corresponding laser point according to the millisecond timestamp, and then the vector addition operation is performed on the original coordinates to obtain the corrected three-dimensional coordinate point set. For example, point (1.500, 0.200, 0.100) plus the compensation amount (0.042, -0.005, 0.021) to get the correction point This process eliminates spatial distortion caused by vehicle motion, such as Z-axis compression caused by vehicle head tilt on uphill sections, which causes the ballast height in the original point cloud to be underestimated by 2.1 cm. This allows the corrected point cloud to restore the true surface morphology.

[0081] Then, in step 1023, the millimeter wave radar echo data is parsed into the dielectric constant characteristic value of the ballast layer and mapped to the corrected three-dimensional coordinate point set output in step 1022 according to the spatial position. Depth layer is divided by time gating technology: According to the electromagnetic wave propagation formula Reverse dielectric constant , where c is the speed of light, t is the echo time, and d is the depth. is the relative dielectric constant, reflecting the dielectric properties of the ballast material, such as dry crushed stone ≈ 3, and water-containing compacted material > 12. For example, (1.542, 0.195, 0.121) corresponds to the area 20 cm underground. The measured echo time is t=1.35ns, and the calculated value is , indicating the presence of water-containing crust at that location. Finally, the dielectric constant is bound to the spatial coordinate point as a physical property.

[0082] Finally, step 1024 is used to expand the attribute dimension of each 3D coordinate point in the correction 3D coordinate point set, creating a composite data unit containing both spatial coordinates and dielectric constant values. (1.542, 0.195, 0.121) and dielectric constant =12.8 is fused into a four-dimensional data structure (1.542, 0.195, 0.121, 12.8). All composite units are aggregated to form an enhanced point cloud model. Its spatial distribution inherits the millimeter-level geometric accuracy of the laser point cloud, and can distinguish the undulation of 2mm ballast particles. The physical dimension carries the dielectric characteristics acquired by radar penetration, such as the dielectric constant value of dry crushed stone. ≈3.0, the dielectric constant value of the saturated junction region >12.0. This model realizes the integrated characterization of ballast's "surface morphology and internal properties", providing a multimodal data foundation for subsequent intelligent identification.

[0083] In practical applications, during inspections on the E Railway, when the inspection vehicle passed through the F curve section at 28 km / h, the IMU recorded a longitudinal displacement of +0.8m, a lateral offset of -0.05m, and a pitch angle of -1.2°. A compensation vector (+0.011, -0.003, -0.028) was calculated for an original laser point (5.214, -1.032, 0.208), and the corrected coordinates were updated to (5.225, -1.035, 0.180). The synchronous millimeter-wave radar echo analyzed the dielectric constant of the layer 20cm deep directly below the point. =14.7, corresponding to the water-bearing compacted material, and finally a composite data unit (5.225, -1.035, 0.180, 14.7) is generated, which together with the surrounding 3,200 similar units constitute the enhanced point cloud model of this section.

[0084] The 102-point overall solution eliminates vehicle motion distortion through dynamic compensation, ensuring point cloud geometric accuracy. By integrating dielectric constant characteristics, the point cloud is upgraded from a purely geometric model to a physical model that integrates both form and quality. The resulting enhanced point cloud model accurately reflects the true morphology of the ballast surface while providing a perspective view of the internal material properties, providing a high-information data foundation for subsequent intelligent recognition.

[0085] 103. Perform a spiral scan on the surface of the ballast area using a line scanning laser module at the bottom of the track inspection vehicle, and synchronously trigger the millimeter-wave radar to perform a penetrating scan to generate a three-dimensional spatial data matrix that integrates the surface topography and internal dielectric characteristics;

[0086] Optionally, step 103 may specifically include the following steps:

[0087] 1031. Based on the forward direction, the galvanometer deflection assembly of the line scanning laser module at the bottom of the track inspection vehicle is controlled to move along the Archimedean spiral trajectory, driving the laser beam to cover the surface of the ballast area in a spiral path, and generating a three-dimensional surface coordinate point set of the ballast surface topography by receiving the reflected laser beam;

[0088] 1032. Send a trigger pulse to the millimeter-wave radar controller at the start of the laser beam scanning to drive the millimeter-wave radar to emit a penetrating beam. After receiving the reflected signal from the ballast layer, the dielectric constant characteristic values ​​at different depths of the ballast layer are extracted.

[0089] 1033. Based on the surface three-dimensional coordinate point set, vertically project the surface coordinate points to the underground coordinate system, and establish a spatial mapping relationship between the surface three-dimensional coordinate points and the underground depth layers according to the dielectric constant characteristic values ​​of the depth layers below the projection points according to the projection positions;

[0090] 1034. Integrate data according to the spatial grid based on the spatial mapping relationship, store surface height values ​​and corresponding underground dielectric constant profile values ​​with the surface projection point as the center position, and aggregate all spatial grid units to generate a three-dimensional spatial data matrix.

[0091] In the above scheme, the galvanometer deflection assembly is the core optical actuator that controls the scanning direction of the laser beam, used to precisely guide the laser beam to the target surface. The Archimedean spiral trajectory refers to the spatial scanning path formed by the laser beam on the ballast surface. Its mathematical characteristic is that the radial distance expands linearly with the rotation angle, achieving complete coverage from the center to the periphery. The surface three-dimensional coordinate point set refers to the spatial position data set of the ballast surface generated by laser ranging. It contains the horizontal coordinates (x, y) and vertical height value z of each scanning point and is used to characterize the geometric distribution of ballast particles. The penetrating beam is the electromagnetic signal emitted by the millimeter-wave radar. Its physical property is that it can penetrate the ballast gravel layer and reflect at the interface between different media. It is used to obtain the dielectric properties of the internal structure of the ballast. The underground coordinate system is a three-dimensional spatial reference system extending vertically downward from the surface reference plane as its origin. Its depth axis 𝑧 is zeroed at the surface and is used to locate the spatial position of underground dielectric features. The dielectric constant profile value refers to the series of dielectric constant data corresponding to different depths below the same surface projection point, reflecting the variation of the physical properties of the ballast layer along the depth direction. A three-dimensional spatial data matrix is ​​a structured data container organized according to a regular spatial grid. Each grid cell stores the surface height value and the associated underground dielectric constant profile.

[0092] In the embodiment of the present application, first, step 1031 performs surface scanning by the line scanning laser module at the bottom of the rail inspection vehicle, and the galvanometer deflection assembly is controlled to move along the Archimedean spiral trajectory, and its path is driven by the preset angular velocity and radial expansion velocity to calculate the coordinates of the laser landing point. When the laser beam covers the ballast surface in a spiral path, the receiver records the time difference and angle of the reflected signal, and combines it with high-precision inertial navigation data to generate a three-dimensional coordinate point set of the ballast surface topography. For example, at t=0.5s, the scanning point coordinates are calculated. , and the height of the point z = 0.15m is measured, forming the surface point (-0.1, 0.0, 0.15).

[0093] Secondly, at the start of the laser scan, a hardware-level synchronous trigger pulse is sent to the millimeter-wave radar controller through step 1032 to drive the radar to emit a 77GHz frequency-modulated continuous wave to penetrate the ballast layer. When the electromagnetic wave encounters the interface of different media and produces reflections, such as the gap between gravel or the plate, and the receiver captures the echo signal, the dielectric constant characteristic value of each depth layer is extracted through the time domain analysis algorithm. For example, at the surface point (-0.1, 0.0), the radar detects that the echo time delay is 4.2ns at 30cm underground. According to the formula The calculated dielectric constant is 15.3, at this time c=3e8m / s, d=0.3m, which represents the core area of ​​the hardening.

[0094] Next, in step 1033, the surface three-dimensional coordinate point set generated in step 1031 is vertically projected to the underground coordinate system: (8.732, 1.204, 0.21) as an example, its projection reference point is set as (8.732,1.204,0). Divide the depth layer downward from the center, with each 10cm as a layer, and bind the dielectric constant characteristic value obtained in 1032 to the corresponding layer according to the depth. The depth sequence below the point [0-10cm, 10-20cm, 20-30cm] is associated with the dielectric value [3.1, 5.2, 15.3], forming a vertical mapping chain of "surface position-subsurface physical properties".

[0095] Finally, in step 1034, the data is gridded based on the spatial mapping relationship, dividing the detection area into 0.1m×0.1m surface grid cells. Each cell integrates the mean height of all surface points within it and the associated underground dielectric constant profile value. For example, the grid (87,12) contains three surface points, with a calculated mean height of 0.22m and a bounded dielectric constant profile of [3.1, 5.2, 15.3]. Finally, a 200×150 grid cell matrix is ​​aggregated to generate a three-dimensional spatial data matrix, where the grid (90,15) stores the data {Height: 0.18m, Dielectric Profile: [3.0, 4.1, 8.9, 15.2]}, which intuitively reveals the distribution of the 1.2m deep hardening zone.

[0096] In actual applications, when the track inspection vehicle travels at a speed of 20 km / h, the galvanometer deflection assembly first drives the laser beam to scan along the Archimedean spiral trajectory at an angular velocity of 8πrad / s and a radial velocity of 0.15m / s. At t=0.6s, a surface point (2.874, -0.532, 0.18) is generated. At the same time, a pulse is triggered to start the millimeter-wave radar. At this point, an echo delay of 3.8ns is detected at a depth of 25cm underground, and the dielectric constant is calculated. =14.2; then the surface point is vertically projected to the grid (28, -5) and associated with the dielectric profile value [3.2, 4.5, 14.2, 12.1] at a depth of 0-40 cm below it; finally, 500 grids in the region are integrated to generate a three-dimensional data matrix, in which the grid (30, -6) stores the mean surface height of 0.21 m and the dielectric profile [3.0, 4.0, 8.5, 15.0], accurately locating the core area of ​​the compaction at a depth of 35 cm.

[0097] The 103 overall solution achieves comprehensive coverage of the ballast surface through spiral scanning, while simultaneously performing penetrating scans to capture internal dielectric characteristics and establish a precise surface-to-subsurface spatial mapping. The resulting three-dimensional spatial data matrix integrates surface topography and internal physical properties within a unified spatial framework, forming a fully integrated "surface-to-subsurface" data base, providing spatially aligned multi-source information for subsequent intelligent recognition.

[0098] 104. Using a neural network structure to analyze the curvature variation law data of the contact surface of the ballast particles in the three-dimensional spatial data matrix and identify the spatial distribution characteristics of the gaps between the ballast particles, while dynamically enhancing the weight of the loose area through a spatiotemporal attention mechanism;

[0099] Optionally, step 104 may specifically include the following steps:

[0100] 1041. Process the three-dimensional spatial data matrix using a neural network structure, extract a surface normal vector change sequence of the ballast particle contact surface region in a primary processing module of the neural network structure, calculate a curvature change quantization value based on the surface normal vector change sequence, and generate data on a curvature change pattern of the ballast particle contact surface;

[0101] 1042. Detecting, in the advanced processing module of the neural network structure, a continuous spatial region having a constant curvature quantization value within the three-dimensional spatial data matrix, identifying the continuous spatial region as a ballast particle gap set, measuring a three-dimensional spatial distribution parameter of the ballast particle gap set, and generating a ballast particle gap spatial distribution feature;

[0102] 1043. Constructing a spatiotemporal attention mechanism based on the spatial distribution characteristics of ballast particle gaps;

[0103] 1044. Increase the weight adjustment increment for the gap area between ballast particles with high frequency spatial position changes, and apply the time accumulation enhancement factor to the gap area between ballast particles that continues to expand over time;

[0104] 1045. Increase the weight value of the loose area in the spatial distribution feature based on the weight adjustment increment and the time accumulation enhancement factor.

[0105] In the above scheme, the surface normal vector change sequence refers to dynamic data describing the geometric evolution of the ballast particle contact surface. It includes a sequence of angular differences between perpendicular vectors at adjacent surface points and can be used to quantify surface concave and convex features. The curvature change quantification value is a numerical indicator that characterizes the sharpness of ballast particle edges and corners and can be used to identify stress concentration areas on the particle contact surface. The ballast particle gap set refers to a three-dimensional continuous spatial region identified as unfilled and can be used to assess the structural stability of the ballast layer. The three-dimensional spatial distribution parameter is a set of quantitative indicators describing the geometric characteristics of the particle gaps and can be used to determine the gap morphology type. The spatiotemporal attention mechanism is an intelligent algorithm that dynamically adjusts data weights. It includes a spatial position offset calculation module and a temporal evolution trend analysis module and can be used to focus on high-risk loose areas. The weight adjustment increment is a reinforcement coefficient applied to areas with high spatial frequency changes and can be used to increase the detection priority of sudden changes. The time accumulation enhancement factor is a progressive reinforcement coefficient applied to continuously expanding areas and can be used to warn of progressive loosening.

[0106] In the embodiment of the present application, the three-dimensional spatial data matrix is ​​first input into the primary processing module of the 3D convolutional neural network through step 1041. The surface elevation data is slidably scanned by a 5×5×5 convolution kernel to calculate the normal vector of each point, that is, the direction vector perpendicular to the surface. Taking the point (1.2, 0.5, 0.18) as an example, the normal vector sequence within its neighborhood of 3 cm is extracted [(0.12, 0.85, 0.51), (0.15, 0.82, 0.54)], and the rate of change of the angle between adjacent normal vectors is calculated: ,in is the curvature quantization value, is the normal vector of the adjacent point, The distance between adjacent points on the surface is calculated by traversing all contact surface areas to generate a matrix of curvature change law data. The curvature of a certain angular area suddenly increases to 0.31mm⁻¹, while the normal area is less than 0.05mm⁻¹.

[0107] Next, in step 1042, the curvature data is input into the graph neural network advanced processing module, which automatically detects continuous regions with curvature values ​​less than 0.01 mm⁻¹. These continuous spatial regions are identified as a set of ballast particle gaps and determined as gaps. A three-dimensional region growing algorithm is executed with the coordinates (120.5, 8.7) as the center, identifying a connected space with a volume of 3.2 m³. Volume parameters are calculated by voxel count and resolution conversion. The surface area is 15.6 m² calculated by integrating the surface triangles. The axial extreme value difference is measured to obtain a depth of 3.2 m and a width of 1.5 m, resulting in an aspect ratio of 2.1. Finally, the feature vector of this gap is output as [position (120.5, 8.7), volume 3.2, surface area 15.6, aspect ratio 2.1] as the spatial distribution characteristics of the ballast particle gaps.

[0108] Next, in step 1043, a dual-channel data structure is constructed based on the spatial distribution characteristics of the ballast particle gaps. The spatial channel loads the coordinates of all current gap centers, for example, establishing a location set L = {(120.5, 8.7, 0.2), (122.1, 9.3, 0.3)}. The temporal channel associates the historical database and extracts the volume sequence T = {2.8, 3.0, 3.2} m³ for the past three measurements at that location. A spatiotemporal encoder binds the location coordinates to historical trends to form a timestamped feature map, establishing a data base for subsequent dynamic analysis.

[0109] Then, step 1044 is used to implement the enhancement strategy for the two risk areas: first, the spatial position offset is calculated, and when the Euclidean distance between the current gap center and the position of the previous week exceeds the threshold of 0.2m, the spatial weight increment is triggered. Secondly, the time trend is analyzed. When the volume value increases by 6.7% and 7.1% for three consecutive times and both exceed the 5% threshold, a time accumulation factor is applied according to the cycle number exponentially. For example, the Euclidean distance of 0.28m between the current gap center (120.5, 8.7) and the last week's position (120.2, 8.9) exceeds the 0.2m threshold, triggering a spatial weight increment. When the volume increases by 6.7% and 7.1% for three consecutive times and both exceed the 5% threshold, a time accumulation factor is applied exponentially according to the number of cycles. At this time, the gap satisfies both the conditions of spatial mutation and continuous expansion in time.

[0110] Finally, the final risk weight value is synthesized through step 1045: take the basic weight , multiplied by the spatial weight increment and time accumulation factor , get the weight value of the loose area When the weight value is greater than 2.0, it is automatically marked as a high-risk loose area, and the gap is given a red warning sign. At the same time, the latest volume is added to the historical sequence for trend analysis in the next cycle, completing the dynamic monitoring closed loop. For example, the basic weight of the gap (120.5, 8.7) is , then the spatial weight increment is =1.5, the time accumulation factor is =1.728, the weight value of the loose area is calculated , at this time, the weight value is greater than 2.0, and it is automatically marked as a high-risk loose area and given a red warning sign. The latest volume 3.2m³ is appended to the historical sequence and updated to T={3.0,3.2,3.2}.

[0111] In practical applications, in the N section of the M Railway, the neural network processes the three-dimensional data matrix: the primary module calculates the normal vector sequence [(0.10, 0.88, 0.46), (0.08, 0.90, 0.42)] for the neighborhood of point (120.5, 8.7, 0.23). The dot product of 0.978 arc cosine yields 12.1°, which, combined with the surface distance of 5.8 mm, yields a curvature of 0.035 mm. - ¹, while the curvature of the adjacent corner area jumps to 0.33mm - ¹; The advanced module identifies a region of constant curvature of 0.008 mm at (120.5, 8.7) - ¹, the gap volume is measured to be 3.2m³ and the surface area is 15.6m² after 3D segmentation; the spatiotemporal mechanism loads the historical volume sequence of this point [2.8, 3.0, 3.2]m³ and the position of last week (120.2, 8.9), and calculates the spatial offset 0.28m trigger increment =1.5, volume continuous growth (7.1%, 6.7%) trigger time factor =1.2³=1.728; the final weight value of the loose area is 1.0×1.5×1.728=2.592>2.0, which is marked as a high-risk loose area and the historical database is updated.

[0112] The above-mentioned 104 overall solution uses a neural network to accurately quantify the curvature changes and gap spatial distribution of the ballast particle contact surface, breaking through the limitations of traditional manual experience; it combines the spatiotemporal attention mechanism to dynamically track position offsets and continuously expanding high-risk areas, and assigns weight increments and cumulative factors; finally, it automatically marks loose risk areas based on the synthetic weight, realizing a closed loop from micro-feature extraction to macro-risk warning, and providing a data-driven dynamic decision-making basis for ballast structure stability assessment.

[0113] 105. Construct a multi-scale point cloud generation model based on an adversarial training mechanism, and geometrically complete the missing occluded areas in the spatial distribution features to output a virtual point cloud of the complete ballast accumulation state;

[0114] Optionally, step 105 may specifically include the following steps:

[0115] 1051. Construct a multi-scale point cloud generation model based on an adversarial training mechanism, comprising a generator unit and a discriminator unit, wherein the generator unit is provided with a coarse-grained generation channel and a fine-grained generation channel, and the discriminator unit is provided with a geometric continuity verification module;

[0116] 1052. Input the occlusion area in the spatial distribution feature into the generator unit, predict the basic geometry of the occlusion area through the coarse-grained generation channel, and simultaneously add the surface details of the ballast particles through the fine-grained generation channel to generate supplementary point cloud data of the occlusion area;

[0117] 1053. The supplementary point cloud data and the original spatial distribution features are input into the discriminator unit. The curvature connection state of the supplementary geometric body and the adjacent area is detected by the geometric continuity verification module, and the physical rationality of the particle gap transition is verified, and the discrimination result is output.

[0118] 1054. When the discrimination result fails to pass the verification, the generator unit parameters are iteratively optimized to regenerate the supplementary point cloud data. When the discrimination result passes the verification, the supplementary point cloud data is spliced ​​with the original spatial distribution characteristics to form a three-dimensional point set data of the complete ballast accumulation state as a virtual point cloud output.

[0119] In the above scheme, the generator unit refers to an occlusion region prediction module based on an adversarial training mechanism. It comprises a dual-channel collaborative structure of coarse-grained geometric framework generation and fine-grained surface detail addition, and is used to output physically plausible supplementary point cloud data. The coarse-grained generation channel is a neural network component that generates basic geometric outlines and is used to construct a macroscopic 3D framework of the occluded region. The fine-grained generation channel is a neural network component that restores the microscopic features of the ballast and is used to add granular texture fluctuations. The discriminator unit is an intelligent evaluation module that verifies the rationality of the generated results. It includes dual detection logic for geometric continuity and physical laws, which is used to ensure the engineering reliability of the completed point cloud. The geometric continuity verification module is an algorithm unit that detects the quality of morphological connections and is used to eliminate geometric discontinuities. The physical plausibility verification module is a logic unit that constrains mechanical compliance, including gap ratio threshold determination, to ensure that the generated results comply with ballast structural mechanics specifications. The supplementary point cloud data refers to the virtual 3D point set generated for the occluded region. It consists of coarse-grained basic framework points and fine-grained surface detail points and is used to reconstruct the complete ballast accumulation state.

[0120] In this embodiment, step 1051 first constructs a dual-module structure based on an adversarial training framework. The generator unit uses the PointNet++ network architecture, where the coarse-grained generation channel is designed as a fully connected layer outputting 256 base points to construct a macro-geometric framework, and the fine-grained generation channel uses a 3D transposed convolutional layer to output 2048 detail points to add granular texture. The discriminator unit integrates a geometric continuity verification module and a physical rationality verification module. The network weights are initialized based on ballast engineering specifications. The coarse-grained channel learns ballast pile slope characteristics, such as a 25° inclination, while the fine-grained channel learns the standard deviation of the gravel surface, such as ±2cm undulation.

[0121] Secondly, the occlusion area boundary point set is input into the generator through step 1052: the coarse-grained channel fits the plane equation through the least squares method to generate grid basic points; the fine-grained channel adds random perturbations based on the generated coordinates , simulate the characteristics of ballast particles. For example, the input occlusion area boundary is {(10.2,3.1,0.18),(10.3,3.2,0.20)}, the coarse-grained channel generates a 20×20 grid base point according to the plane equation z=0.25x+0.1y, and the fine-grained channel adds random perturbations , simulating the characteristics of ballast particles to form surface concave and convex details.

[0122] Next, the supplemented point cloud is merged with the original point cloud in step 1053 and then input into the discriminator: the geometric continuity module calculates the curvature of the edge points of the supplemented area , and the curvature of the adjacent original points The difference is worth it ,when It passes the verification when it is less than the threshold of 0.02; then the total volume of the generated area is calculated through the physical rationality module , gap volume , calculate the gap ratio and compare it with the threshold of 40% to determine whether it meets the ballast mechanics specifications. The calculation formula is as follows: For example, in the decision maker, the geometric continuity module calculates the curvature of the edge points of the completion area. Curvature of adjacent original points The difference The value is less than the threshold of 0.02 and passes the verification; at the same time, the physical rationality module calculates the total volume of the generated area. , gap volume The calculated gap ratio is 35% < the threshold value of 40%, which is judged to be in compliance with the ballast mechanics specifications.

[0123] Finally, iterative optimization and stitching are performed in step 1054. If verification passes, the supplementary points are directly spliced ​​with the original points. If verification fails, the generator parameters are adjusted using the backpropagation algorithm: the learning rate of the coarse-grained channel is reduced to 0.0001 to smooth the geometric framework, and the noise amplitude of the fine-grained channel is increased to ±3cm to enhance detail diversity. The supplementary point cloud is regenerated until it passes verification, and finally a complete virtual point cloud is output.

[0124] In practical applications, in the detection of the S section of the R railway, the boundary point set {(15.2, 4.3, 0.22), (15.3, 4.4, 0.25)} is input into the occluded area; the coarse-grained channel of the generator fits the plane equation z=0.28x+0.12y to generate an 18×18 basic grid, and the fine-grained channel superimposes the perturbation function Convert the basic point (15.25, 4.35, 0.255) to a fine-grained point (15.25, 4.35, 0.268); the discriminator detects the completed edge curvature of 0.072mm - ¹0.069mm from the original area - The difference of 0.003 between the two values ​​was less than the threshold of 0.02. Furthermore, the gap volume of 0.38 m³ / total volume of 1.05 m³ resulted in a gap ratio of 36.2% less than the 40% threshold. Finally, the 1,650 supplementary points were combined with the original 7,800 points to produce a complete virtual point cloud. When the gap ratio reached 42% in the initial generation, the discriminator triggered parameter optimization. After adjusting the fine-grained noise amplitude, the second generation achieved a gap ratio of 37.1%, passing verification.

[0125] The overall solution of 105 described above achieves intelligent completion of occluded areas through adversarial training: the coarse-grained channel ensures macroscopic geometric rationality, and the fine-grained channel restores microscopic particle characteristics. The dual verification of the discriminator ensures that the completion result is both morphologically consistent and meets engineering mechanics constraints. The final output of the complete virtual point cloud provides a reliable data foundation for quantitative analysis of ballast status.

[0126] 106. Quantitatively analyze the virtual point cloud, combine the dielectric constant distribution characteristics, the curvature variation law data, and the loose area weight in the enhanced point cloud model, and output the ballast density value and the spatial coordinate range of the compacted area, which serve as the key identification result of the environmental status of the ballast structure along the railway.

[0127] Optionally, step 106 may specifically include the following steps:

[0128] 1061. Perform quantitative analysis on the virtual point cloud, calculate the point density distribution of the virtual point cloud within the unit volume spatial grid, and use the median value of the point density distribution as the ballast density reference value;

[0129] 1062. Establish a hardened area determination condition based on the dielectric constant distribution characteristics, curvature variation law data, and loose area weight value in the enhanced point cloud model. The hardened area determination condition includes three independent determination conditions: a first determination condition is to mark areas in the enhanced point cloud model where the dielectric constant characteristic value is continuously higher than a set threshold; a second determination condition is to extract flat areas in the curvature variation law data where the curvature variation is lower than the curvature threshold; and a third determination condition is to locate areas in the loose area weight value distribution where the weight value is lower than a critical weight value.

[0130] 1063. Determine that a region that satisfies the first, second, and third determination conditions is a hardened region, extract the coordinates of the spatial turning points of the outer contour of the hardened region, and connect the coordinates of the spatial turning points to form a three-dimensional polygonal bounding box.

[0131] 1064. The ballast density reference value is output as the ballast density value, and the vertex coordinate set of the three-dimensional polygonal bounding box is output as the spatial coordinate range of the compacted area. The ballast density value and the spatial coordinate range of the compacted area together constitute the key identification result of the environmental status of the ballast structure along the railway.

[0132] In the above scheme, point density distribution refers to a quantitative indicator reflecting the density of ballast, including the statistical characteristics of the number of virtual point clouds within a unit volume spatial grid, and can be used to map the ballast density benchmark value. The ballast density benchmark value refers to a normalized value that characterizes the overall density state of the ballast, including a scaled value linearly converted through the median of the point density, and can be used to evaluate the compactness of the ballast layer structure. The compaction area determination condition refers to a triple logical constraint for identifying the compaction characteristics of ballast, including a synergistic mechanism of dielectric constant threshold, curvature flatness threshold, and weight critical value, which can be used to accurately locate the compaction area. The three-dimensional polygonal bounding box refers to a geometric structure that describes the spatial extent of the compaction area, including a set of convex polyhedron vertices generated by connecting the spatial turning points of the outer contour, and can be used to define the machine-readable boundary of the compaction area. The spatial turning point coordinates refer to the key position points on the boundary of the compaction area, including the horizontal projection polygon corner points and the depth extension extreme points, and can be used to construct the geometric framework of the three-dimensional bounding box.

[0133] In the embodiment of the present application, the complete virtual point cloud is first divided into a spatial grid in step 1061, and a 0.1m³ cubic unit is used to cover the detection area, and the number of point clouds contained in each unit is counted. The formula for calculating the unit volume point density is: , where N is the number of points and V=0.001m³. Collect all grid density values ​​and sort them to get the median , which is converted into a density benchmark value through linear mapping to represent the overall density of the ballast. The conversion formula is as follows: ,in Points / m³ is a theoretical maximum. For example, a grid (50,60) contains 158 points, resulting in a calculated density of 158,000 points / m³. The median density of 500 grid cells is 152,000 points / m³, giving a baseline density of 0.76.

[0134] Secondly, in step 1062, the three types of data are coordinated to establish a logical AND gate criterion: first, the dielectric distribution of the enhanced point cloud is scanned and the dielectric constants of more than three consecutive grid cells are marked. >12; secondly, extract the curvature data with curvature change <0.05mm - Finally, locate low-risk cells with loose weight values ​​<1.0. Only when the spatial location meets three independent conditions at the same time is marked as a candidate for plate knotting area. For example, the dielectric value of 14.2>12 at the coordinate (120.5,8.7) and the curvature change of 0.04mm - ¹<0.05, and the loose weight value 0.9<1.0, are marked as candidate plate-bonded units.

[0135] Next, perform 3D boundary extraction on the candidate plate knot area in step 1063: use edge detection algorithm to identify the peripheral corner points of the horizontal projection of the area, and determine the depth range by combining the dielectric anomaly peak. A convex hull algorithm is used to connect spatial turning points: Horizontally, corner points are connected to form a polygon, and vertically, they are extended to the depth extremes to generate a minimum volume 3D bounding box. For example, the corner points (120.3, 8.6) (120.7, 8.6) (120.7, 8.9) (120.3, 8.9) and the depth 0.3-0.6m generate an 8-vertex bounding box {(120.3, 8.6, 0.3)...(120.3, 8.9, 0.6)}.

[0136] Finally, in step 1064, the analysis results are integrated. The density baseline value obtained in step 1061 is used as the ballast density value. The vertex coordinates of the three-dimensional bounding box generated in step 1063 are used as the spatial extent of the compacted area. This is then packaged into key-value pairs in a machine-readable format, creating a machine-parseable railway ballast condition diagnosis report that provides data support for maintenance decisions. For example, the key identification results can be packaged in JSON format: {"Compactness": 0.76,"Compacted Area": ​​[[120.3, 8.6, 0.3], [120.7, 8.6, 0.3], ...]}.

[0137] In practical application, in the detection of the W section of the V railway, the virtual point cloud was divided into 0.1m³ grid units. The grid (60,70) counted 162 points with a density of 162,000 points / m³. The median density of 500 grids was 154,000 points / m³, and the density benchmark value was 0.77 after linear mapping. The coordinate (130.2,9.5) was marked as a candidate area for compaction because it met the three conditions of dielectric value 15.1>12 threshold, curvature 0.03mm⁻¹<0.05 threshold, and weight 0.85<1.0 threshold. The peripheral corner points (130.0 ,9.3),(130.4,9.3),(130.4,9.7),(130.0,9.7) and depth 0.4-0.7m, generate 8-vertex bounding box {(130.0,9.3,0.4)..(130.0,9.7,0.7)} through convex hull algorithm; finally output structured result {"density":0.77,"hardening area":[[130.0,9.3,0.4],[130.4,9.3,0.4],...[130.0,9.7,0.7]]}, completing the complete diagnosis from data quantification to spatial positioning.

[0138] The above-mentioned 106 overall solution quantifies the overall compaction state of ballast through point density distribution, breaking through the limitations of traditional qualitative assessment. It accurately identifies the spatial extent of the compacted area based on the coordinated judgment of multiple conditions such as dielectric characteristics, curvature flatness, and weight values. The final output structured data provides a locatable and quantifiable decision-making basis for railway maintenance.

[0139] The following is a complete example for steps 101 to 106. Figure 2As shown in the figure, when the track inspection vehicle enters the Y section at a speed of 25 km / h, the laser scanning and millimeter-wave radar are started simultaneously: the line scanning laser module collects the point cloud of the ballast surface at a frequency of 200 Hz, and generates 12,500 three-dimensional coordinate points in a single frame; at the same time, the IMU records the vehicle speed of 25 km / h and the pitch angle of -0.8°; the millimeter-wave radar detects the dielectric constant 20 cm below the coordinates (85.32, 4.15). =14.7 is an abnormally high value. The three types of data are precisely aligned using the unified timestamp "2023-09-15 10:30:25.456".

[0140] Then, the laser point cloud is dynamically compensated based on the IMU state parameters: the pitch compensation ΔZ=0.018m is calculated for the point (85.32, 4.15, 0.21), and the corrected coordinates are (85.32, 4.15, 0.192); the radar dielectric value =14.7 and this point is bound to generate a composite data unit (85.32, 4.15, 0.192, 14.7). 8,200 enhanced point cloud units are constructed for the entire area, of which the point (86.10, 4.22, 0.180, 3.2) represents dry gravel.

[0141] Next, the spiral scan and penetration scan fusion are started: the laser beam scans the surface along an Archimedean spiral trajectory at an angular velocity of 6πrad / s, generating a point (85.32, 4.15, 0.192) at t=0.7s; the radar is synchronously triggered to detect the dielectric profile [3.1, 4.5, 14.7] (0-30cm) below this point; the data {height 0.192m, dielectric profile [3.1, 4.5, 14.7]} is projected onto the grid (853, 415) to store data, forming a three-dimensional spatial matrix of 200×150 grids.

[0142] Then, the neural network analyzes the structural features: the primary module calculates the neighborhood curvature of the point (85.32, 4.15) as 0.35mm⁻¹; the advanced module identifies a gap set with a volume of 2.8m³ at (85.32, 4.15); the spatiotemporal attention mechanism monitors that the gap has shifted by 0.22m compared to last week and its volume has continuously increased by 11%. It applies a spatial increment of 1.5 and a temporal factor of 1.2³=1.728, and the combined weight rises to 2.592, marking a high-risk loose area.

[0143] Adversarial generative model to complete occluded areas: input occluded boundary points , the coarse-grained channel generates a 15×15 basic grid; the fine-grained channel adds perturbations The point (85.32, 4.12, 0.19) is converted to (85.32, 4.12, 0.198); the discriminator verifies that the curvature difference 0.012 is less than the 0.02 threshold and the gap rate 37% is less than 40%, and outputs 1,500 supplementary points to be spliced ​​with the original 9,800 points to form a complete virtual point cloud.

[0144] Finally, quantitative analysis and result output: The virtual point cloud is divided into grids, and the median density is 154,000 points / m³, which is mapped to a density of 0.77; (85.32, 4.15) satisfies both =14.7>12, curvature 0.04<0.05, weight 0.85<1.0; extract corner points (85.30, 4.10), (85.35, 4.10), (85.35, 4.20), (85.30, 4.20) and generate an 8-vertex bounding box at a depth of 0.3-0.6m. The final structured output is: {"Compactness": 0.77, "Crusted Area": ​​[[85.30, 4.10, 0.3], [85.35, 4.10, 0.3], … [85.30, 4.20, 0.6]]}.

[0145] Figure 3 The present invention provides a schematic diagram of a railway environment recognition system based on artificial intelligence. Figure 3 As shown, the system includes:

[0146] Acquisition module 31, used to acquire laser point cloud data of the track ballast area, and synchronously collect the running status parameters of the track inspection vehicle and the echo data of the millimeter wave radar;

[0147] A construction module 32 is configured to perform dynamic distortion compensation on the laser point cloud data based on the operating state parameters, and to construct an enhanced point cloud model including physical properties in combination with dielectric constant characteristics in the echo data;

[0148] A generation module 33 is configured to perform a spiral scan of the surface of the ballast area using a line scanning laser module at the bottom of the track inspection vehicle, and synchronously trigger the millimeter-wave radar to perform a penetrating scan to generate a three-dimensional spatial data matrix that integrates the surface topography and internal dielectric characteristics;

[0149] An analysis module 34 is configured to analyze the curvature variation law data of the contact surface of the ballast particles in the three-dimensional spatial data matrix using a neural network structure and identify the spatial distribution characteristics of the gaps between the ballast particles, while dynamically enhancing the weight of loose areas through a spatiotemporal attention mechanism;

[0150] A completion module 35 is configured to construct a multi-scale point cloud generation model based on an adversarial training mechanism, and geometrically complete the missing occluded areas in the spatial distribution features to output a virtual point cloud of a complete ballast accumulation state;

[0151] The output module 36 is used to perform quantitative analysis on the virtual point cloud, combine the dielectric constant distribution characteristics, the curvature change law data and the loose area weight in the enhanced point cloud model, output the ballast density value and the spatial coordinate range of the compacted area, and use it as the key identification result of the environmental status of the ballast structure along the railway.

[0152] Figure 3 The railway environment recognition system based on artificial intelligence can be executed Figure 1 The implementation principles and technical effects of the artificial intelligence-based railway environment recognition method described in the illustrated embodiment will not be elaborated on here. The specific manner in which each module and unit performs operations in the artificial intelligence-based railway environment recognition system in the above embodiment has been described in detail in the embodiments of the method and will not be elaborated on here.

[0153] In one possible design, Figure 3 The railway environment recognition system based on artificial intelligence in the embodiment shown can be implemented as a computing device, such as Figure 4 As shown, the computing device may include a storage component 41 and a processing component 42;

[0154] The storage component 41 stores one or more computer instructions, wherein the one or more computer instructions are called and executed by the processing component 42 .

[0155] The processing component 42 is used for the above Figure 1 The embodiment provides an artificial intelligence-based method for identifying railway environment.

[0156] The processing component 42 may include one or more processors to execute computer instructions to complete all or part of the steps in the above method. Of course, the processing component may also be implemented as one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to perform the above method.

[0157] The storage component 41 is configured to store various types of data to support operations at the terminal. The storage component can be implemented by any type of volatile or non-volatile memory device, or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk, or optical disk.

[0158] Of course, a computing device may also include other components, such as input / output interfaces, display components, communication components, etc.

[0159] The input / output interface provides an interface between the processing component and the peripheral interface module, which can be an output device, an input device, etc.

[0160] The communication component is configured to facilitate, among other things, wired or wireless communications between the computing device and other devices.

[0161] Among them, the computing device can be a physical device or an elastic computing host provided by a cloud computing platform, etc. In this case, the computing device can refer to a cloud server, and the above-mentioned processing components, storage components, etc. can be basic server resources rented or purchased from the cloud computing platform.

[0162] The present application also provides a computer storage medium storing a computer program, wherein the computer program can achieve the above-mentioned Figure 1 The illustrated embodiment is an artificial intelligence-based method for identifying railway environment.

[0163] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0164] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, i.e., they may be located in one location or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of the present embodiment. Persons of ordinary skill in the art will be able to understand and implement the present invention without inventive effort.

[0165] Through the above description of the embodiments, those skilled in the art will clearly understand that each embodiment can be implemented using software plus a necessary general-purpose hardware platform, or of course, hardware. Based on this understanding, the essence of the above technical solution, or the portion that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, or an optical disk, and includes a number of instructions for causing a computer device (such as a personal computer or server) to execute the methods described in each embodiment or certain portions of the embodiments.

[0166] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for identifying railway environment based on artificial intelligence, characterized in that: include: Acquire laser point cloud data of the track ballast area, and simultaneously collect the operating status parameters of the track inspection vehicle and the echo data of the millimeter-wave radar; Performing dynamic distortion compensation on the laser point cloud data based on the operating state parameters, and constructing an enhanced point cloud model including physical properties in combination with dielectric constant characteristics in the echo data; The surface of the ballast area is spirally scanned by a line-scan laser module at the bottom of the track inspection vehicle, and the millimeter-wave radar is synchronously triggered to perform a penetrating scan to generate a three-dimensional spatial data matrix that integrates the surface topography and internal dielectric characteristics; A neural network structure is used to analyze the curvature variation law data of the ballast particle contact surface in the three-dimensional spatial data matrix and identify the spatial distribution characteristics of the gaps between the ballast particles. At the same time, the weight of the loose area is dynamically enhanced through a spatiotemporal attention mechanism. A multi-scale point cloud generation model is constructed based on an adversarial training mechanism to geometrically complete the missing occluded areas in the spatial distribution features to output a virtual point cloud of the complete ballast accumulation state; The virtual point cloud is quantitatively analyzed, and combined with the dielectric constant distribution characteristics, the curvature change law data and the loose area weight in the enhanced point cloud model, the ballast density value and the spatial coordinate range of the compacted area are output, which serve as the key identification results of the environmental status of the ballast structure along the railway.

2. The method according to claim 1, characterized in that The multi-scale point cloud generation model is constructed based on the adversarial training mechanism, and the missing occluded areas in the spatial distribution features are geometrically completed to output a virtual point cloud of a complete ballast accumulation state, including: Based on the adversarial training mechanism, a multi-scale point cloud generation model is constructed, which includes a generator unit and a discriminator unit. The generator unit is provided with a coarse-grained generation channel and a fine-grained generation channel, and the discriminator unit is provided with a geometric continuity verification module. Inputting the occlusion area in the spatial distribution feature into the generator unit, predicting the basic geometry of the occlusion area through the coarse-grained generation channel, and adding the surface details of the ballast particles through the fine-grained generation channel to generate supplementary point cloud data of the occlusion area; The supplementary point cloud data and the original spatial distribution features are input into the discriminator unit, and the curvature connection state of the supplementary geometric body and the adjacent area is detected by the geometric continuity verification module, while the physical rationality of the particle gap transition is verified, and the discrimination result is output; When the discrimination result fails to pass the verification, the generator unit parameters are iteratively optimized to regenerate supplementary point cloud data. When the discrimination result passes the verification, the supplementary point cloud data is spliced ​​with the original spatial distribution characteristics to form a three-dimensional point set data of the complete ballast accumulation state as a virtual point cloud output.

3. The method according to claim 1, characterized in that The dynamic enhancement of loose region weights through the spatiotemporal attention mechanism includes: A spatiotemporal attention mechanism is constructed based on the spatial distribution characteristics of the gaps between ballast particles; Add weight adjustment increments to the gap areas between ballast particles with high frequency spatial position changes, and apply time accumulation enhancement factors to the gap areas between ballast particles that continue to expand over time; The weight value of the loose area in the spatial distribution feature is increased based on the weight adjustment increment and the time accumulation enhancement factor.

4. The method according to claim 1, wherein The quantitative analysis of the virtual point cloud is combined with the dielectric constant distribution characteristics, the curvature change law data and the loose area weight in the enhanced point cloud model to output the ballast density value and the spatial coordinate range of the compacted area, which are used as key identification results of the environmental status of the ballast structure along the railway, including: Performing quantitative analysis on the virtual point cloud, calculating the point density distribution of the virtual point cloud within a unit volume spatial grid, and using the median value of the point density distribution as a ballast density reference value; Combined with the dielectric constant distribution characteristics, curvature change law data and loose area weight value in the enhanced point cloud model, a hardened area determination condition is established. The hardened area determination condition includes three independent determination conditions: the first determination condition is to mark the area in the enhanced point cloud model where the dielectric constant characteristic value is continuously higher than the set threshold; the second determination condition is to extract the flat area where the curvature change amount is lower than the curvature threshold in the curvature change law data; the third determination condition is to locate the area with a weight value lower than the critical weight value in the loose area weight value distribution; Determine an area that satisfies the first, second, and third determination conditions simultaneously as a hardened area, extract the coordinates of spatial turning points of the outer contour of the hardened area, and connect the coordinates of the spatial turning points to form a three-dimensional polygonal bounding box; The ballast density reference value is output as the ballast density value, and the vertex coordinate set of the three-dimensional polygonal bounding box is output as the spatial coordinate range of the compacted area. The ballast density value and the spatial coordinate range of the compacted area together constitute the key identification result of the environmental status of the ballast structure along the railway.

5. The method according to claim 1, wherein The line scanning laser module at the bottom of the track inspection vehicle performs a spiral scan on the surface of the ballast area, and synchronously triggers the millimeter wave radar to perform a penetration scan to generate a three-dimensional spatial data matrix that integrates the surface topography and internal dielectric characteristics, including: Based on the forward direction control, the galvanometer deflection assembly of the line scanning laser module at the bottom of the track inspection vehicle moves along the Archimedean spiral trajectory, driving the laser beam to cover the surface of the ballast area in a spiral path, and generating a three-dimensional coordinate point set of the ballast surface topography by receiving the reflected laser beam; At the start of the laser beam scan, a trigger pulse is sent to the millimeter-wave radar controller to drive the millimeter-wave radar to emit a penetrating beam. After receiving the reflected signal from the ballast layer, the dielectric constant characteristic values ​​at different depths of the ballast layer are extracted. Based on the surface three-dimensional coordinate point set, the surface coordinate point is vertically projected onto the underground coordinate system, and the dielectric constant characteristic value of each depth layer below the projection point is associated with the projection position to establish a spatial mapping relationship between the surface three-dimensional coordinate point and the underground depth layer; Based on the spatial mapping relationship, the data is integrated into a spatial grid, the surface height value and the corresponding underground dielectric constant profile value are stored with the surface projection point as the center position, and all spatial grid units are aggregated to generate a three-dimensional spatial data matrix.

6. The method according to claim 1, characterized in that The method of using a neural network structure to analyze the curvature variation regularity data of the contact surface of the ballast particles in the three-dimensional spatial data matrix and identifying the spatial distribution characteristics of the gaps between the ballast particles includes: Processing the three-dimensional spatial data matrix through a neural network structure, extracting a surface normal vector change sequence of the ballast particle contact surface area in a primary processing module of the neural network structure, calculating a curvature change quantization value based on the surface normal vector change sequence, and generating data on the curvature change pattern of the ballast particle contact surface; In the advanced processing module of the neural network structure, a continuous spatial region with a constant curvature quantization value in the three-dimensional spatial data matrix is ​​detected, the continuous spatial region is identified as a set of ballast particle gaps, the three-dimensional spatial distribution parameters of the set of ballast particle gaps are measured, and the spatial distribution characteristics of the ballast particle gaps are generated.

7. The method according to claim 1, characterized in that The dynamically compensating the laser point cloud data for distortion based on the operating state parameters and constructing an enhanced point cloud model including physical properties in combination with the dielectric constant characteristics in the echo data includes: Generate a position compensation vector for each three-dimensional coordinate point in the laser point cloud data based on the longitudinal displacement, lateral offset and pitch angle values ​​in the operating state parameters; According to the position compensation vector, the original coordinates of the laser point cloud data are subjected to point-by-point translation correction to obtain a corrected three-dimensional coordinate point set; Analyzing the dielectric constant characteristic values ​​at different depths of the ballast layer from the millimeter-wave radar echo data, and mapping the dielectric constant characteristic values ​​as new physical properties to corresponding points in the corrected three-dimensional coordinate point set according to spatial positions; The attribute dimension is expanded for each corrected 3D coordinate point to generate a composite data unit containing both spatial coordinates and dielectric constant values. All composite data units are aggregated to form an enhanced point cloud model.

8. A railway environment recognition system based on artificial intelligence, characterized in that: include: The acquisition module is used to obtain laser point cloud data of the track ballast area and synchronously collect the operating status parameters of the track inspection vehicle and the echo data of the millimeter wave radar; A construction module, configured to perform dynamic distortion compensation on the laser point cloud data based on the operating state parameters, and to construct an enhanced point cloud model including physical properties in combination with dielectric constant characteristics in the echo data; A generation module is used to perform a spiral scan on the surface of the ballast area using a line scanning laser module at the bottom of the track inspection vehicle, and synchronously trigger the millimeter wave radar to perform a penetration scan to generate a three-dimensional spatial data matrix that integrates the surface topography and internal dielectric characteristics; an analysis module for analyzing the curvature variation law data of the contact surface of the ballast particles in the three-dimensional spatial data matrix using a neural network structure and identifying the spatial distribution characteristics of the gaps between the ballast particles, while dynamically enhancing the weight of loose areas through a spatiotemporal attention mechanism; A completion module is used to build a multi-scale point cloud generation model based on an adversarial training mechanism, and geometrically complete the missing occluded areas in the spatial distribution features to output a virtual point cloud of the complete ballast accumulation state; The output module is used to perform quantitative analysis on the virtual point cloud, combine the dielectric constant distribution characteristics in the enhanced point cloud model, the curvature change law data and the loose area weight, output the ballast density value and the spatial coordinate range of the compacted area, and use it as the key identification result of the environmental status of the ballast structure along the railway.

9. A computing device, characterized in that It comprises a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are used to be called and executed by the processing component to implement an artificial intelligence-based railway environment identification method as described in any one of claims 1 to 7.

10. A computer storage medium, characterized in that A computer program is stored, and when the computer program is executed by a computer, the method for identifying the railway environment based on artificial intelligence as described in any one of claims 1 to 7 is implemented.

Images