Train carriage hidden injury identification method and system based on enhanced infrared imaging

Abstract

The invention belongs to the technical field of non-destructive detection, and particularly relates to a train carriage hidden injury identification method and system based on enhanced infrared imaging. The method comprises the following steps: collecting a visible light image and an infrared image of each carriage of a train; obtaining codes and sequence data of the carriage based on the visible light image; based on the visible light image, the infrared image and the coding and sequence data of the carriage, identifying the key part to obtain a region identification infrared image; carrying out preprocessing based on the region identification infrared image to obtain an enhanced infrared image; and carrying out hidden injury identification on the enhanced infrared image by adopting a pre-constructed separation model, judging whether a hidden injury exists or not, and determining the position size of the hidden injury. The problems that an existing method cannot track, detect and recognize a moving target and recognition is inaccurate are solved.

Description

Technical Field

[0001] The present invention belongs to the technical field of nondestructive detection, and in particular relates to a method and system for identifying hidden damage in a train carriage based on enhanced infrared imaging. Background Art

[0002] Coal transport trains may be affected by various factors during long-term use, resulting in the breakage and cracking of key components, such as the breakage of the middle door column and the breakage of the small door belt. If not discovered in time, there will be great safety hazards, causing unnecessary personnel and economic losses. Therefore, relevant hidden damage detection methods are needed to monitor the trains in real time, so as to timely discover hidden damage problems and take relevant measures in time to improve the safety of the coal transportation process.

[0003] As a flexible and fast general technology, non-destructive testing and evaluation technology is widely used in many fields such as aerospace, military, and construction. As an interdisciplinary and cross-application field of general practical technology, infrared thermal imaging is an effective alternative and supplement to traditional non-destructive testing technology. Infrared thermal imaging is a non-destructive testing method based on the principle of infrared radiation. It can analyze the surface and internal defects or structures of the sample by scanning, recording or observing the temperature changes of the surface being tested. Compared with traditional non-destructive testing such as ultrasound and X-ray, infrared thermal imaging has the advantages of fast measurement speed, intuitive measurement results, large detection area, and easy automation. However, it is difficult to directly obtain defect information through direct infrared detection. External energy excitation is required to make the surface defects of the object to be tested and its surrounding environment produce radiation differences in order to effectively distinguish. This type of excitation usually includes laser, ultrasound, high-frequency thermal pulses, etc. Photothermal pulses are also a major auxiliary excitation imaging method. As a type of auxiliary excitation infrared thermal imaging, pulse thermal imaging uses a photothermal pulse light source to illuminate the target surface and quickly heat the surface of the object. When the heat wave is transmitted to the internal defects of the specimen, it is blocked and propagates in the opposite direction, so that a temperature difference will be generated between the surface corresponding to the defective part and the surrounding non-defective part. The internal defects of the target object can be detected by using an infrared thermal imager to detect the temperature distribution on the target surface. However, it has not yet been used in the real-time monitoring and identification of hidden damage in moving train carriages.

[0004] However, due to the need for detection under the condition of carriage movement during actual on-site maintenance and inspection, the existing visible light-infrared detection methods are unable to track, detect and identify moving targets. In addition, existing methods for hidden damage to train carriages include artificial intelligence technology and manual methods. Artificial intelligence technology builds an image acquisition system on both sides of the train track, collects train images and transmits them to the computing platform for calculation, and identifies problem vehicles. However, the camera acquisition system may not be able to collect problem images because the broken parts are covered by foam glue or coal powder, resulting in the failure to identify problem vehicles, affecting train driving safety; the manual method is to establish inspection stations and manually inspect key parts of the vehicle, which requires a lot of manpower and has many problems such as high labor intensity and long time for workers. Summary of the invention

[0005] The purpose of the present invention is to provide a method and system for identifying hidden damage in train carriages based on enhanced infrared imaging, so as to solve the problem that existing methods cannot track, detect and identify moving targets and the identification is inaccurate.

[0006] The present invention is achieved through the following technical solutions: A method for identifying hidden damage in a train carriage based on enhanced infrared imaging comprises the following steps: S1, collect visible light images and infrared images of each carriage of the train; S2, obtaining the encoding and sequence data of the carriage based on the visible light image; S3, based on the visible light image, infrared image and the coding and sequence data of the carriage, identify the key parts and obtain the regional identification infrared image; S4, preprocessing the infrared image based on the region recognition to obtain an enhanced infrared image; S5. Use the pre-built segmentation model to identify hidden damage in the enhanced infrared image, determine whether there is hidden damage and determine the location and size of the hidden damage.

[0007] Furthermore, before S1, it is determined in advance whether there is an oncoming vehicle. If there is no oncoming vehicle, the detection is suspended; If there is an approaching vehicle, collect the temperature and humidity of the environment; Determine whether the weather condition is severe weather. If so, suspend detection; if not, start S1.

[0008] Further, S3 is specifically: The target tracking algorithm is used to extract the key parts of the visible light image and obtain the extraction frame; Based on the coding and sequence data of the carriage, the extraction frame is matched with the infrared image corresponding to the extraction frame to obtain a region recognition infrared image; The key parts are easily damaged parts.

[0009] Further, in S4, the preprocessing comprises the following steps: S4.1, based on the region recognition infrared image, an adaptive piecewise linear transformation method is used in the spatial domain to obtain the gray value of each pixel after transformation, thereby forming the infrared image matrix data; S4.2, performing a center-symmetric transformation on the infrared image matrix data to obtain a center-transformed infrared image; S4.3, performing discrete Fourier transform on the infrared image after the center transformation according to the Fourier transform function, so that the infrared image after the center transformation is converted from the spatial domain to the frequency domain; S4.4, in the frequency domain, using the transfer function of a Gaussian high-pass filter to filter the infrared image to obtain a filtered infrared image; S4.5. The filtered infrared image is converted from the frequency domain to the spatial domain through inverse Fourier transform, and a central symmetric transformation is performed to obtain an enhanced infrared image.

[0010] Further, in S5, the construction process of the pre-constructed segmentation model is: The DyHead dynamic object detection head is introduced after the decoder of the original DeepLabV3+ deep learning image segmentation model to obtain an improved segmentation model; Collecting historical data and establishing sample data; the historical data includes visible light images and enhanced infrared images of train carriages; Divide the sample data into training set and test set; The improved segmentation model is trained using the training set, and the trained segmentation model is tested using the test set until the test meets the requirements, thereby obtaining the pre-constructed segmentation model.

[0011] The present invention also discloses a train carriage hidden damage identification system based on enhanced infrared imaging, comprising: Infrared light source module, carriage information collection module and hidden damage feature recognition module; the carriage information collection module is connected with the hidden damage feature recognition module; The hidden damage feature recognition module includes a visible light feature recognition module and an infrared feature recognition module; the infrared feature recognition module includes a hidden damage detection target tracking module, a data preprocessing module and a hidden damage recognition module; The carriage information acquisition module is used to collect visible light images of each carriage of the train, and obtain infrared images of each carriage with the assistance of the infrared light source module; A visible light feature recognition module is used to perform visible light feature recognition on the visible light image to obtain the code and sequence data of the carriage; Hidden damage detection target tracking module, used to identify key parts based on visible light images, infrared images and the coding and sequence data of the carriage, and obtain regional identification infrared images; A data preprocessing module performs preprocessing on the infrared image based on the region recognition to obtain an enhanced infrared image; The hidden damage recognition module is used to identify hidden damage in the enhanced infrared image using a pre-built segmentation model, determine whether there is a hidden damage and determine the location and size of the hidden damage.

[0012] Furthermore, the visible light feature recognition module includes a carriage code recognition module, a carriage sequence recognition module and a carriage video storage module; The carriage code recognition module is used to recognize the carriage code in the visible light image based on the video stream template matching algorithm to obtain the carriage code data; The carriage sequence recognition module is used to determine the carriage sequence number based on the semantic pulse frequency of the character code and obtain the sequence data of the carriage; The carriage video storage module is used to store the data of each carriage, and the data of each carriage includes the visible light image, coding data and sequence data of each carriage.

[0013] Furthermore, infrared light source modules are deployed on both sides of the train track to generate infrared pulse radiation to enhance the infrared image information collected by the carriage information collection module; The carriage information collection module includes a pole bracket, a monitoring PTZ, a visible light camera and an infrared camera; Pole brackets are deployed on both sides of the train track, and a monitoring gimbal is mounted on the pole brackets to carry visible light cameras and infrared cameras; the visible light camera is used to collect visible light images of each carriage, and the infrared camera is used to obtain enhanced infrared images of each carriage with the assistance of an infrared light source module.

[0014] Furthermore, the train carriage hidden damage identification system further includes an incoming vehicle monitoring module, an environment identification module and a control module; the incoming vehicle monitoring module and the environment identification module are both connected to the control module; The vehicle monitoring module is used to monitor whether there is an oncoming vehicle. If there is an oncoming vehicle, the control module starts the environment recognition module; The environment recognition module is used to collect the temperature and humidity of the environment, and judge whether the weather condition is bad weather based on the temperature and humidity. If it is bad weather, the detection is suspended; if it is not bad weather, the control module controls the infrared light source module and the car information collection module to work.

[0015] Further, the vehicle approaching monitoring module includes a track clamp, a screw rod, an L-shaped bracket and an active magnetic steel; The track clamp is clamped on the track through a screw rod, the L-shaped bracket is fixed on the track clamp, and the active magnetic steel is fixed on the L-shaped bracket.

[0016] Compared with the prior art, the present invention has the following beneficial technical effects: The present invention discloses a method for identifying hidden damage in train carriages based on enhanced infrared imaging. First, visible light images and infrared images are collected. The rich color, texture and other detailed information contained in the visible light images and the characteristics of the infrared images that can reflect the temperature distribution and potential defects of objects are used to provide multi-dimensional data support for subsequent comprehensive analysis, so as to more comprehensively understand the condition of the carriages and improve the accuracy and reliability of hidden damage identification. By performing feature recognition on the visible light images to obtain the coding and sequence data of the carriages, the carriages can be accurately identified and sorted, which is convenient for the subsequent management and analysis of the data of each carriage, providing a comprehensive understanding of the entire detection process. Clear carriage information records help to quickly locate and track the detection situation of a specific carriage, and improve the management efficiency of the detection work; before the hidden damage is identified, the present invention also pre-processes the regional identification infrared image, which can improve the quality of the infrared image and obtain an enhanced infrared image. While achieving the goals of reducing image noise and improving image contrast, it also solves the problem of possible blurring of crack edges in the image, and improves the accuracy of subsequent crack geometric feature extraction; finally, a pre-built segmentation model is used to highlight the features related to hidden damage and suppress irrelevant information, so as to more accurately determine whether there is a hidden damage and the location and size of the hidden damage. This recognition method of the present invention can improve the accuracy and reliability of hidden damage identification and reduce misjudgment and missed judgment.

[0017] Furthermore, before collecting visible light images and enhanced infrared images, it is first determined whether to start the subsequent detection process by judging whether there is an oncoming vehicle, so as to avoid unnecessary detection operations when there is no vehicle, save system resources, improve the operating efficiency of the detection system, ensure that the detection process is started at the right time, and make the detection work proceed in an orderly manner; and also consider the impact of ambient temperature and humidity and weather conditions on the detection. Severe weather (such as heavy rain, heavy fog, etc.) may affect the quality of visible light images and infrared images, resulting in inaccurate collected data or inability to accurately identify hidden damage; when it is severe weather, suspending the detection can avoid erroneous detection results due to interference from severe weather, ensure the validity of the detection data and the reliability of the detection results, and reduce misjudgments and missed judgments.

[0018] Furthermore, the target tracking algorithm is used to extract the key parts and obtain the extraction frame. The target tracking algorithm can accurately locate and track the key parts of the carriage, quickly find the area of ​​interest in complex image data, and improve the accuracy and speed of key part recognition.

[0019] Furthermore, the present invention first uses an adaptive piecewise linear transformation method to process the image, reduce image noise and improve image contrast, but the crack edge in the image is blurred, which will affect the accuracy of the subsequent crack geometric feature extraction; in order to better solve this problem, the present invention sharpens the bottom crack infrared image enhanced by the adaptive piecewise linear transformation method. A new infrared image enhancement algorithm combining adaptive piecewise linear transformation with Gaussian high-pass filtering is used. The algorithm enhances the contrast of the door crack infrared image in the spatial domain, reduces image noise, and sharpens the crack infrared image in the frequency domain, enhancing the detailed information of the steel defect part, solving the problem of blurring the crack edge in the image directly using the adaptive piecewise linear transformation method, which affects the accuracy of the crack geometric feature extraction.

[0020] Furthermore, the present invention improves the DeepLabV3+ deep learning image segmentation model, introduces the DyHead dynamic target detection head, and obtains an improved segmentation model. This method integrates multiple attention mechanisms to unify scale perception, spatial perception, and task perception in target detection, significantly improving the context perception ability of the network, and thus improving the representation ability of the detection head. Through continuous training and testing, the parameters of the model are adjusted so that the model can better adapt to the task of identifying hidden damage in the carriage, and improve the accuracy and reliability of the model. When the test meets the requirements, the obtained segmentation model can effectively identify hidden damage in train carriages in practical applications.

[0021] The invention discloses a train carriage hidden damage identification system based on enhanced infrared imaging, comprising an infrared light source module, a carriage information acquisition module and a hidden damage feature recognition module. The carriage information acquisition module can capture images from moving carriages in real time, and collect visible light images and infrared images of each carriage of the train; the visible light image and the infrared image are comprehensively used to obtain multi-dimensional information of the carriage, the visible light image provides rich details such as color and texture, and the infrared image can reflect the temperature distribution and potential defects. The combination of the two makes the detection more comprehensive, can capture more hidden damage clues, and improves the accuracy of the detection; the visible light feature recognition module encodes and sequentially recognizes the carriages, and saves relevant data, which is convenient for managing and tracing the detection information of each carriage, can quickly locate the detection situation of a specific carriage, and provide convenience for subsequent analysis and decision-making, which is helpful to improve the management level of the detection work; then the hidden damage detection target tracking module identifies the key parts and obtains the regional recognition infrared image; the data preprocessing module preprocesses the regional recognition infrared image to obtain the enhanced infrared image; the hidden damage recognition module is used to identify the hidden damage of the enhanced infrared image, so as to realize the automation and intelligence of the detection process. Compared with manual inspection, it greatly improves the inspection efficiency, reduces the misjudgment and missed judgment caused by human factors, and can also quickly process a large amount of carriage inspection data. This system can not only determine whether there are hidden damages in the carriage, but also determine the location and size of the hidden damages. These detailed information provides maintenance personnel with clear maintenance guidance, making maintenance work more targeted, and can reasonably arrange maintenance resources, improve maintenance efficiency, and reduce maintenance costs.

[0022] Furthermore, the visible light feature recognition module includes a car code recognition module, a car sequence recognition module and a car video storage module; the car code recognition module determines the approximate location of the car code area based on the car type, determines the specific code area based on the previous and next frame information obtained from the video stream as a template, removes interference from other character areas in the car body, and then can recognize the car code character function based on the character recognition algorithm; the car sequence recognition module can determine the car sequence number based on the semantic pulse frequency of the character code; the car video storage module saves the data of each car based on the number. This effectively realizes the acquisition of basic car information, and when hidden damage is detected, the specific car location of the hidden damage can be immediately obtained, so that the problem car can be promptly checked.

[0023] Furthermore, the carriage information collection module uses a monitoring pan-tilt dual-view probe that combines an infrared camera and a visible light camera to detect hidden damage in key parts of the train carriage, such as pillars and belts. It can capture and process images from moving carriages in real time, and combined with a matching target tracking algorithm, it can obtain detection image information of key areas, thereby effectively identifying potential hidden damage.

[0024] Furthermore, the incoming train monitoring module can accurately determine the arrival and departure of trains, start and stop the detection program in time, so that the entire detection process can proceed in an orderly manner, avoiding invalid detection when there is no train, saving system resources, and improving the operating efficiency of the detection system; the environmental recognition module determines the weather and environmental conditions by collecting the current ambient temperature and humidity. In bad weather, the control car information collection module suspends detection to avoid inaccurate detection results due to interference from environmental factors, thereby ensuring the validity and reliability of the detection data. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 This is a schematic diagram of the hardware structure of a train carriage hidden damage identification system based on enhanced infrared imaging according to the present invention; Figure 2 This is a module connection diagram of a train carriage hidden damage identification system based on enhanced infrared imaging according to the present invention; Figure 3 This is a structural diagram of the vehicle monitoring module; Figure 4 This is the flow chart of the hidden injury feature recognition module; Figure 5 For the improved partition model; Figure 6 This is a flow chart of a method for identifying hidden damage in a train carriage based on enhanced infrared imaging according to the present invention; Figure 7 For Figure 6 A flow chart of a train carriage hidden damage identification method based on enhanced infrared imaging optimized on the basis of the present invention; Among them, 1. Carriage information collection module; 2. Infrared light source module; 3. Hidden damage feature recognition module; 4. Oncoming vehicle monitoring module; 5. Environment recognition module; 6. Control module; 101. Pole bracket; 102. Monitoring pan / tilt; 103. Visible light camera; 104. Infrared camera; 201, bracket; 202, infrared pulse light source; 203, spotlight cover; 401, track clamp; 402, screw; 403, L-shaped bracket; 404, active magnetic steel. DETAILED DESCRIPTION

[0026] In order to make the purpose, technical solution and advantages of the present invention more clear, the following is further described in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention, that is, the embodiments described are only part of the embodiments of the present invention, not all embodiments.

[0027] The components described and shown in the drawings and embodiments of the present invention may be arranged and designed in various configurations. Therefore, the detailed description of the embodiments of the present invention provided in the following drawings is not intended to limit the scope of the claimed invention, but merely represents a selected embodiment of the present invention. Based on the drawings and embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0028] It should be noted that the terms "comprises", "includes" or any other variations are intended to cover non-exclusive inclusion, so that a process, element, method, article or equipment that includes a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to the process, element, method, article or equipment.

[0029] The features and performances of the present invention are further described in detail below in conjunction with the embodiments.

[0030] Example 1 like Figure 1 and Figure 2 As shown, the present invention discloses a train carriage hidden damage identification system based on enhanced infrared imaging, including a carriage information acquisition module 1, an infrared light source module 2 and a hidden damage feature identification module 3; the carriage information acquisition module 1 is connected to the hidden damage feature identification module 3.

[0031] The infrared light source module 2 is installed on both sides of the train's traveling direction, and is used to generate high-energy infrared pulse radiation to instantly generate pulse heat on the surface of the tested carriage, so as to enhance the infrared image information collected by the carriage information collection module 1 and enhance the infrared display of hidden damage inside the carriage structure.

[0032] The carriage information acquisition module 1 is used to obtain visible light images and infrared images of the train carriages, realize visible and infrared data collection of key areas such as pillars and belts of each carriage, and transmit the information to the hidden damage feature recognition module 3.

[0033] The hidden damage feature recognition module 3 is used to perform specific area tracking detection and recognition on the information collected by the carriage information collection module 1, and to perform hidden damage feature recognition.

[0034] Example 2 Based on Example 1, the specific structures of the carriage information acquisition module 1 and the infrared light source module 2 are introduced.

[0035] like Figure 1As shown, the carriage information acquisition module 1 includes a pole bracket 101, a monitoring platform 102, a visible light camera 103 and an infrared camera 104. The pole bracket 101 is deployed on both sides of the train track, and the monitoring platform 102 is mounted on the pole bracket 101 to carry the visible light camera 103 and the infrared camera 104. The visible light camera 103 is used to collect visible light images of the carriage, and the infrared camera 104 is used to obtain enhanced infrared images of each carriage under the auxiliary enhancement of the infrared light source module 2.

[0036] The infrared light source module 2 includes a bracket 201, an infrared pulse light source 202 and a spotlight cover 203. The bracket 201 is deployed on both sides of the train track, the infrared pulse light source 202 is located on the bracket 201, and is used to generate high-energy infrared pulse radiation, and the spotlight cover 203 is installed on the infrared pulse light source 202, and is used to focus and protect the infrared pulse light source 202.

[0037] The infrared pulse light source 202 is a plurality of lamps arranged in parallel longitudinally, and the pulse frequency of the infrared light source is adjustable within a range of 1-50 Hz.

[0038] The infrared pulse light source 202 can quickly heat the surface of an object by releasing a large amount of energy instantly to generate high-intensity light pulses. This fast and uniform thermal excitation method can induce a temperature gradient inside the material, thereby helping to detect tiny structural defects, delamination, cracks, etc. Since the infrared pulse duration is extremely short, usually in milliseconds or shorter, it can avoid long-term thermal effects on the sample and is suitable for the detection of sensitive materials. In addition, the infrared pulse light source 202 is highly controllable and can adjust the energy output and pulse frequency according to different detection requirements. The infrared pulse light source 202 is arranged in a multi-lamp longitudinal parallel manner at intervals of 0.65 m, and the pulse frequency is adjustable from 1 to 50 Hz.

[0039] When the vehicle detector ahead detects an approaching vehicle, if there is no approaching vehicle, it will continue to be in standby mode; if there is an approaching vehicle, the weather conditions will be determined. If it is bad weather, in order to ensure the safety and reliability of the infrared light source module 2, the detection will be suspended to protect the dual-view probe; if it is normal weather, the monitoring will continue; the train carriage moves forward at a constant speed along the detection line, the infrared pulse light source 202 and the acquisition device are controlled to work, and the enhanced excitation is started. At the same time, the visible light camera 103 and the infrared camera 104 collect data from each carriage at a certain frequency. When it is detected that the train has left, the infrared pulse light source 202 and the dual-view probe are turned off, and the standby mode is returned. The collected data is transmitted to the monitoring data center in real time. Based on the dual-view image data, the hidden damage feature recognition module 3 is used to identify whether there are hidden damages in the target area according to the train movement rhythm. Generate a vehicle inspection report, save the hidden damage data, and upload it to the user interface.

[0040] Example 3 On the basis of Example 1, Figure 2 As shown, the present invention discloses a train carriage hidden damage identification system based on enhanced infrared imaging, which also includes an incoming vehicle monitoring module 4, an environment identification module 5 and a control module 6. The carriage information acquisition module 1, the infrared light source module 2, the incoming vehicle monitoring module 4 and the environment identification module 5 are connected to the control module 6 respectively.

[0041] The vehicle monitoring module 4 is used to detect whether a train enters the detection area and transmit the information to the control module 6. After determining that there is an oncoming vehicle, the control module 6 starts the detection program. After the vehicle is offline, the control module 6 stops the detection program. The environment recognition module 5 is used to determine the current environmental conditions. The environment recognition module 5 determines the weather and environmental conditions by collecting the current environmental temperature and humidity and transmits the information to the control module 6. In bad weather, the control module 6 controls the cabin information collection module to suspend detection to ensure the safety of the infrared light source module 2.

[0042] like Figure 3 As shown, the vehicle monitoring module 4 includes a track fixture 401, a screw 402 of the track fixture 401, an L-shaped bracket 403, and an active magnetic steel 404. The track fixture 401 is deployed on the track, and the screw 402 is used to fasten the track fixture 401 and the track, and the L-shaped bracket 403 and the track. The active magnetic steel 404 is connected to the L-shaped bracket 403. When a vehicle approaches, the active magnetic steel 404 will output a pulse to transmit the vehicle information to the control module 6 in advance to start the pre-inspection program.

[0043] The vehicle monitoring module 4 is deployed in front of and behind the vehicle compartment information collection module 1. When a vehicle is detected passing by, it determines that the vehicle is offline and transmits the information to the control module 6, and the detection program stops.

[0044] The two active magnetic steels 404 are named A and B and are briefly described as A and B below.

[0045] Installation method: A and B are installed at a certain distance (such as 3-5 meters) along the track direction to cover the same track section.

[0046] The working principle of the vehicle monitoring module 4 is: When a train comes, the front wheels trigger A first, and then the wheels continue to move forward to trigger B. When the sequential signals of A→B are detected, it is determined that the train has entered the detection area.

[0047] After the train has passed all the wheels, A and B have no new trigger signals, and the trigger times match. If there is no new signal for a preset time after the last trigger of B, the train is considered to have passed all the wheels.

[0048] The environment identification module 5 includes a temperature and humidity detection device, which transmits the temperature and humidity detection information to the control module 6 to determine whether it is bad weather such as rain or snow. If it is not bad weather, the control module 6 controls the subsequent detection process to proceed normally. If so, the protection device is started and the detection is suspended.

[0049] Example 4 Based on Example 1, the composition of the hidden injury feature recognition module 3 is introduced.

[0050] The hidden damage feature recognition module 3 includes a visible light feature recognition module and an infrared feature recognition module. The visible light feature recognition module is used for encoding and identification of carriage numbers, and the data of each carriage is encoded and saved separately in real time according to the identification results; the infrared feature recognition module is used for target area tracking and infrared video segmentation and storage of infrared images.

[0051] like Figure 4 As shown, the basic structure of the visible light feature recognition module is: a car code recognition module, a car sequence recognition module and a car video storage module.

[0052] The car code recognition module realizes car code recognition based on visible light images and video stream template matching algorithm. Its basic process is: first, based on the car type, the approximate location of the car code area is determined, and the specific coding area is determined based on the previous and next frame information obtained from the video stream as a template, the interference of other character areas in the car body is eliminated, and the car code characters are recognized using the character recognition algorithm.

[0053] The carriage sequence recognition module determines the carriage sequence number based on the semantic pulse frequency of the character code.

[0054] The carriage video saving module is used to save the data of each carriage.

[0055] like Figure 4 As shown, the infrared feature recognition module includes a hidden damage detection target tracking module, a data preprocessing module and a hidden damage recognition module.

[0056] The hidden damage detection target tracking module is used to identify key parts such as pillars and belt areas based on visible light images, infrared images, and the coding and sequence data of the carriage to obtain a regional recognition infrared image. Specifically, the target tracking algorithm is used to extract the key parts of the visible light image to obtain an extraction frame; based on the coding and sequence data of the carriage, the extraction frame is matched with the infrared image corresponding to the extraction frame to obtain a regional recognition infrared image.

[0057] The target tracking algorithm is the video stream-based YOLOv5 (You Only Look Once version 5) target detection method. The YOLOv5 target detection method is a single-stage target detection algorithm that converts the target detection task into a regression problem. Its core idea is to divide the input image into multiple grids, each of which is responsible for predicting multiple bounding boxes and their corresponding category probabilities. In the detection based on video streams, the algorithm is applied to the video sequence frame by frame, thereby achieving real-time detection of targets in the video.

[0058] The data preprocessing module is used to preprocess the infrared image based on region recognition to obtain an enhanced infrared image.

[0059] The hidden damage recognition module is used to enhance and extract the enhanced infrared image using a pre-built segmentation model, determine whether there is a hidden damage and determine the location and size of the hidden damage.

[0060] The hidden injury recognition module stores a pre-built segmentation model, which uses an improved DeepLabV3+ deep learning image segmentation model to identify hidden injuries in regional recognition infrared images and determine whether there are hidden injuries and their location and size.

[0061] Example 5 Based on Example 4, the construction process of the partition model is introduced.

[0062] The construction process of the partition model is: like Figure 5 As shown in the figure, based on the original DeepLabV3+ deep learning image segmentation model, the DyHead (Dynamic Head) dynamic target detection head is introduced after the decoder of the model to obtain an improved segmentation model. The introduction of the DyHead dynamic target detection head can improve the context perception ability of the network, thereby improving the accuracy of image segmentation.

[0063] Collect a large amount of historical data and establish sample data; the historical data includes visible light images and enhanced infrared images of train carriages; Divide the sample data into training set and test set; The improved segmentation model is trained using the training set, and the trained segmentation model is tested using the test set until the test meets the requirements and a qualified segmentation model is obtained.

[0064] The DeepLabV3+ deep learning image segmentation model is an advanced image segmentation model that combines the advantages of multi-scale feature fusion and encoder-decoder structure, and can efficiently generate high-resolution segmentation results. Specifically includes an encoder and a decoder. The present invention mainly introduces the DyHead dynamic target detection head in the decoder part. The core idea of ​​the DyHead dynamic target detection head is to regard the input of the target detection head as a 3D tensor with three dimensions of feature level (L) × space (S) × channel (C), and build an attention mechanism on this tensor. This unified detection head can be regarded as an attention learning problem. It is chosen to deploy the attention mechanism separately on each specific dimension of the feature, namely scale-aware attention, space-aware attention, and task-aware attention.

[0065] For the input feature tensor , the attention function of the DyHead dynamic target detection head is specifically: ; in, , , Deployed in , and Attention functions on different dimensions, where is the task-aware attention function, is the spatial perception attention function, is the scale-aware attention function; A feature tensor representing the input F is a three-dimensional tensor; Represents the final result of DyHead output; It is a scale-aware attention function that fuses features of different scales based on their semantic importance: ; in, is a linear function approximated by a 1×1 convolutional layer, It is the hard-sigmoid function.

[0066] It is a spatial perception attention function, focusing on the ability to distinguish different spatial positions: ; in, is the number of coefficient sampling locations, is the spatial offset through self-learning The adjusted position is used to focus on the discrimination area. It's location The self-learning importance scalar at , both of which are learned from the intermediate layer input features; is the index, specifically In the process of summing From 1 to Iteration; refers to the coefficient used to weight the features at a specific location; is the index, specifically In the process of summing From 1 to Iteration.

[0067] It is a task-aware attention function that can jointly learn the generalization of target representations, dynamically turning on / off feature channels to select different tasks: ; in, It is feature slices of channels, is a hyperfunction that learns to control the activation threshold. Implementation: First, dimensionality to reduce the dimensionality, then use two fully connected layers and a normalization layer, and finally apply the offset sigmoid function to normalize the output to within the range.

[0068] Example 6 like Figure 6 As shown, the present invention discloses a method for identifying hidden damage in a train carriage based on enhanced infrared imaging, comprising the following steps: S1, collect visible light images and infrared images of each carriage of the train; S2, the encoding and sequence data of the carriage obtained based on the visible light image; S3, based on the visible light image, infrared image and the coding and sequence data of the carriage, identify the key parts and obtain the regional identification infrared image; S4, preprocessing the infrared image based on the region recognition to obtain an enhanced infrared image; S5. Use the pre-built segmentation model to identify hidden damage in the enhanced infrared image, determine whether there is hidden damage and determine the location and size of the hidden damage.

[0069] Preferably, Figure 7 As shown, before S1, it is pre-determined whether there is an oncoming vehicle. If there is no oncoming vehicle, the detection is suspended; If there is an approaching vehicle, collect the temperature and humidity of the environment; Determine whether the weather condition is severe weather. If so, suspend detection; if not, start S1.

[0070] Example 7 Based on Example 6, S4 is introduced.

[0071] The preprocessing specifically includes the following steps: S4.1. Using the adaptive piecewise linear transformation method in the spatial domain can adaptively enhance different regions according to the grayscale distribution of the image, making the detail areas such as cracks more prominent. The adaptive piecewise transformation is specifically: ; in, ; in, is the pixel coordinate of the image, , Represent the horizontal and vertical coordinates of the pixel position respectively; is the original gray value of each pixel in the infrared image; is the gray value of each pixel after transformation; It is the different intervals of gray level of infrared image; and They are the linear gain factor and offset of each interval, which are adaptively adjusted according to the image features; Represents the number of grayscale intervals of infrared images.

[0072] S4.2, multiply the enhanced infrared image by Perform a central symmetric transformation to obtain an infrared image after central transformation. The central symmetric transformation is specifically as follows: ; in, is the image pixel value after centrosymmetric transformation.

[0073] S4.3, performing discrete Fourier transform on the infrared image after the center transformation according to the Fourier transform function, so that the infrared image after the center transformation is converted from the spatial domain to the frequency domain; The Fourier transform function is specifically: ; in, is the Fourier transform result in the frequency domain, and is the coordinate in the frequency domain, and are the width and height of the infrared image respectively; is a natural exponential function; j represents the imaginary unit; is the ratio of pi.

[0074] S4.4. In the frequency domain, the crack infrared image is filtered using the transfer function of a Gaussian high-pass filter to obtain a filtered infrared image.

[0075] The filtering process allows the information of the edge and internal detail pixels of the image to pass smoothly, while the grayscale distribution of the image will be suppressed, thereby achieving the purpose of enhancing the edge information and detail pixels of the crack image.

[0076] The expression of filtering processing is: ; is the filtered infrared image, Gaussian high-pass filter transfer function; The transfer function of the Gaussian high-pass filter is specifically: ; in, is the distance of each point from the origin in the frequency domain; represents the cutoff frequency; is a natural exponential function; express The square of The transfer function of the Gaussian high-pass filter itself; This represents the transfer function of the Gaussian high-pass filter itself. In the frequency domain, it describes the response characteristics of the filter to different frequency components. Specifically, The value of determines whether the signal component at position (u, v) in the frequency domain can pass through the filter and the degree to which it passes.

[0077] For the center point of the frequency domain , .

[0078] S4.5, convert the filtered infrared image from the frequency domain to the spatial domain through inverse Fourier transform and multiply it by , and get the enhanced infrared image.

[0079] The expression of the inverse Fourier transform is: ; in, is the filtered infrared image, It is an image converted from the frequency domain to the spatial domain through inverse Fourier transform. j represents the imaginary unit; is the circumference of a circle, is a natural exponential function.

[0080] The low-frequency part of the infrared image of the crack in the tested carriage mainly includes the grayscale distribution of the image, and the high-frequency part mainly includes the tiny pixel information of the crack edge and the inside of the image. Directly using the traditional adaptive piecewise linear transformation method to process the image, although the purpose of reducing image noise and improving image contrast is achieved, the crack edge in the image is blurred, which affects the accuracy of the subsequent crack geometric feature extraction. In order to better solve this problem, it is necessary to sharpen the infrared image of the bottom crack enhanced by the adaptive piecewise linear transformation. The processing of the crack infrared image by the Gaussian high-pass filter is in the frequency domain. The information of the edge and internal detail pixels of the image can pass smoothly, while the grayscale distribution of the image will be suppressed, thereby achieving the purpose of enhancing the edge information and detail pixels of the crack image. Since the image sharpened by the Gaussian high-pass filter enhances the image detail information while also increasing the image noise, the image needs to be denoised before sharpening.

[0081] The present invention provides a train carriage hidden damage identification method and system based on enhanced infrared imaging, which are easy to use and have broad application prospects. They can not only be used for monitoring hidden damage of columns and belts in coal transportation train carriages, but can also be expanded to real-time hidden damage detection of other types of trains and even other moving equipment, and have strong promotion value.

[0082] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents, and any modifications or equivalent replacements that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A train carriage hidden damage identification method based on enhanced infrared imaging, characterized in that: The following steps are involved: S1, collect visible light images and infrared images of each carriage of the train; S2, obtaining the encoding and sequence data of the carriage based on the visible light image; S3, based on the visible light image, infrared image and the coding and sequence data of the carriage, identify the key parts and obtain the regional identification infrared image; S4, preprocessing the infrared image based on the region recognition to obtain an enhanced infrared image; S5. Use the pre-built segmentation model to identify hidden damage in the enhanced infrared image, determine whether there is hidden damage and determine the location and size of the hidden damage.

2. According to claim 1, a method for identifying hidden damage in a train carriage based on enhanced infrared imaging is characterized in that: Before S1, determine in advance whether there is an oncoming vehicle. If there is no oncoming vehicle, suspend the detection; If there is an approaching vehicle, collect the temperature and humidity of the environment; Determine whether the weather condition is severe weather. If so, suspend detection; if not, start S1.

3. The method for identifying hidden damage in train carriages based on enhanced infrared imaging according to claim 1 is characterized in that: S3 is specifically: The target tracking algorithm is used to extract the key parts of the visible light image and obtain the extraction frame; Based on the coding and sequence data of the carriage, the extraction frame is matched with the infrared image corresponding to the extraction frame to obtain a region recognition infrared image; The key parts are easily damaged parts.

4. The method for identifying hidden damage in a train carriage based on enhanced infrared imaging according to claim 1 is characterized in that: In S4, the preprocessing comprises the following steps: S4.1, based on the region recognition infrared image, an adaptive piecewise linear transformation method is used in the spatial domain to obtain the gray value of each pixel after transformation, thereby forming the infrared image matrix data; S4.2, performing a center-symmetric transformation on the infrared image matrix data to obtain a center-transformed infrared image; S4.3, performing discrete Fourier transform on the infrared image after the center transformation according to the Fourier transform function, so that the infrared image after the center transformation is converted from the spatial domain to the frequency domain; S4.4, in the frequency domain, using the transfer function of a Gaussian high-pass filter to filter the infrared image to obtain a filtered infrared image; S4.

5. The filtered infrared image is converted from the frequency domain to the spatial domain through inverse Fourier transform, and a central symmetric transformation is performed to obtain an enhanced infrared image.

5. The method for identifying hidden damage in train carriages based on enhanced infrared imaging according to claim 1 is characterized in that: In S5, the construction process of the pre-constructed segmentation model is: The DyHead dynamic object detection head is introduced after the decoder of the original DeepLabV3+ deep learning image segmentation model to obtain an improved segmentation model; Collecting historical data and establishing sample data; the historical data includes visible light images and enhanced infrared images of train carriages; Divide the sample data into training set and test set; The improved segmentation model is trained using the training set, and the trained segmentation model is tested using the test set until the test meets the requirements, thereby obtaining the pre-constructed segmentation model.

6. A train carriage hidden damage identification system based on enhanced infrared imaging, characterized in that: include: Infrared light source module (2), carriage information collection module (1) and hidden damage feature recognition module (3); the carriage information collection module (1) is connected to the hidden damage feature recognition module (3); The hidden damage feature recognition module (3) includes a visible light feature recognition module and an infrared feature recognition module; The infrared feature recognition module includes a hidden damage detection target tracking module, a data preprocessing module and a hidden damage recognition module; A carriage information acquisition module (1) is used to acquire a visible light image of each carriage of a train, and to acquire an infrared image of each carriage with the assistance of an infrared light source module (2); A visible light feature recognition module is used to perform visible light feature recognition on the visible light image to obtain the code and sequence data of the carriage; Hidden damage detection target tracking module, used to identify key parts based on visible light images, infrared images and the coding and sequence data of the carriage, and obtain regional identification infrared images; A data preprocessing module performs preprocessing on the infrared image based on the region recognition to obtain an enhanced infrared image; The hidden damage recognition module is used to identify hidden damage in the enhanced infrared image using a pre-built segmentation model, determine whether there is a hidden damage and determine the location and size of the hidden damage.

7. The train carriage hidden damage identification system based on enhanced infrared imaging according to claim 6 is characterized in that: The visible light feature recognition module includes a carriage code recognition module, a carriage sequence recognition module and a carriage video storage module; The carriage code recognition module is used to recognize the carriage code in the visible light image based on the video stream template matching algorithm to obtain the carriage code data; The carriage sequence recognition module is used to determine the carriage sequence number based on the semantic pulse frequency of the character code and obtain the sequence data of the carriage; The carriage video storage module is used to store the data of each carriage, and the data of each carriage includes the visible light image, coding data and sequence data of each carriage.

8. The train carriage hidden damage identification system based on enhanced infrared imaging according to claim 6 is characterized in that: The infrared light source module (2) is deployed on both sides of the train track to generate infrared pulse radiation to enhance the infrared image information collected by the carriage information collection module (1); The carriage information collection module (1) comprises a pole bracket (101), a monitoring platform (102), a visible light camera (103) and an infrared camera (104); The pole brackets (101) are deployed on both sides of the train track, and the monitoring platform (102) is mounted on the pole brackets (101) and is used to carry a visible light camera (103) and an infrared camera (104); the visible light camera (103) is used to collect visible light images of each carriage, and the infrared camera (104) is used to obtain enhanced infrared images of each carriage under the auxiliary enhancement of the infrared light source module (2).

9. The train carriage hidden damage identification system based on enhanced infrared imaging according to claim 6 is characterized in that: The train carriage hidden damage identification system further comprises an incoming vehicle monitoring module (4), an environment identification module (5) and a control module (6); the incoming vehicle monitoring module (4) and the environment identification module (5) are both connected to the control module (6); The vehicle oncoming monitoring module (4) is used to monitor whether there is an oncoming vehicle. If there is an oncoming vehicle, the control module (6) activates the environment recognition module (5); The environment recognition module (5) is used to collect the temperature and humidity of the environment, and judge whether the weather condition is bad weather according to the temperature and humidity. If it is bad weather, the detection is suspended; if it is not bad weather, the control module (6) controls the infrared light source module (2) and the carriage information collection module (1) to work.

10. The train carriage hidden damage identification system based on enhanced infrared imaging according to claim 9, characterized in that: The vehicle approaching monitoring module (4) comprises a track clamp (401), a screw rod (402), an L-shaped bracket (403) and an active magnetic steel (404); The track clamp (401) is clamped on the track via a screw rod (402), the L-shaped bracket (403) is fixed on the track clamp (401), and the active magnetic steel (404) is fixed on the L-shaped bracket (403).

Images