CN116245875A - Subway train running part temperature detection method and system based on deep learning - Google Patents
Subway train running part temperature detection method and system based on deep learning Download PDFInfo
- Publication number
- CN116245875A CN116245875A CN202310394440.9A CN202310394440A CN116245875A CN 116245875 A CN116245875 A CN 116245875A CN 202310394440 A CN202310394440 A CN 202310394440A CN 116245875 A CN116245875 A CN 116245875A
- Authority
- CN
- China
- Prior art keywords
- heat map
- infrared
- train
- running part
- infrared heat
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000001514 detection method Methods 0.000 title claims abstract description 38
- 238000013135 deep learning Methods 0.000 title claims abstract description 34
- 238000012549 training Methods 0.000 claims abstract description 44
- 230000006870 function Effects 0.000 claims abstract description 22
- 238000000034 method Methods 0.000 claims abstract description 22
- 230000008569 process Effects 0.000 claims abstract description 17
- 238000012545 processing Methods 0.000 claims abstract description 17
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 14
- 238000002372 labelling Methods 0.000 claims abstract description 5
- 238000004519 manufacturing process Methods 0.000 claims description 13
- 230000007246 mechanism Effects 0.000 claims description 12
- 238000004364 calculation method Methods 0.000 claims description 10
- 238000010276 construction Methods 0.000 claims description 7
- 238000004590 computer program Methods 0.000 claims description 6
- 238000001914 filtration Methods 0.000 claims description 5
- 238000009434 installation Methods 0.000 claims description 4
- 238000003860 storage Methods 0.000 claims description 4
- 238000012360 testing method Methods 0.000 claims description 4
- 230000003044 adaptive effect Effects 0.000 claims description 3
- 238000007781 pre-processing Methods 0.000 abstract description 2
- 238000010586 diagram Methods 0.000 description 4
- 238000009529 body temperature measurement Methods 0.000 description 3
- 238000010438 heat treatment Methods 0.000 description 2
- 238000012423 maintenance Methods 0.000 description 2
- 238000005299 abrasion Methods 0.000 description 1
- 230000002411 adverse Effects 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000013145 classification model Methods 0.000 description 1
- 238000000280 densification Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 230000036541 health Effects 0.000 description 1
- 230000004807 localization Effects 0.000 description 1
- 238000005461 lubrication Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000001931 thermography Methods 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0004—Industrial image inspection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/11—Region-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/136—Segmentation; Edge detection involving thresholding
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/25—Determination of region of interest [ROI] or a volume of interest [VOI]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/28—Quantising the image, e.g. histogram thresholding for discrimination between background and foreground patterns
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10024—Color image
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10048—Infrared image
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20092—Interactive image processing based on input by user
- G06T2207/20104—Interactive definition of region of interest [ROI]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30108—Industrial image inspection
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Multimedia (AREA)
- Evolutionary Computation (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Computing Systems (AREA)
- Software Systems (AREA)
- General Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Databases & Information Systems (AREA)
- Quality & Reliability (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Molecular Biology (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Image Processing (AREA)
Abstract
The invention discloses a subway train running part temperature detection method and system based on deep learning, wherein the method comprises the following steps: the infrared image acquisition system acquires an infrared heat map of a target part of the running part in the running process of the train, and performs data labeling and classification label making after preprocessing to obtain an infrared heat map data set; establishing an improved training model of a YOLOv5 deep learning algorithm, integrating a convolution attention module into a backbone network, extracting characteristics of a target component of a running part of a train by using a frame regression loss function with CIoU as a target, and identifying information of the target component of the running part according to the obtained training model of the data set; based on the extracted target component, combining infrared heat map graying treatment to obtain a single-channel gray image, and calculating the temperature value of the target component. The system comprises an infrared image acquisition system, an infrared heat map processing module, a model building module and a temperature calculating module. The invention can accurately and efficiently diagnose the high-temperature fault of key equipment of the running part of the subway train.
Description
Technical Field
The invention relates to the technical field of vehicle bottom target detection and temperature detection, in particular to a subway train running part temperature detection method and system based on deep learning.
Background
The good running state of the running part of the subway train is the guarantee of the safe running of the train, wherein an axle box, a traction motor and a gear box are key equipment of the running part. With the continuous development of subway trains to the running speed and the densification, the phenomenon of heating faults of key components often occurs. Therefore, the reliable detection method is researched, a reasonable running part temperature prediction model is established, and the method has important significance for ensuring running safety and reducing maintenance cost.
Bearings are an integral part of the railway vehicle that is subjected to the highest loads, the health of which directly affects train safety. In the high-speed running process of the train, the bearing can be in a high-load and high-rotation-speed working environment for a long time, and the problems of abrasion, peeling and even breakage of a contact surface can often occur. The traction device mainly comprises a traction motor and a gear box, wherein the traction motor is a power machine for realizing traction and electric braking of a subway train, the mutual conversion of electric energy and mechanical energy is realized, and the torque of the traction motor transmits kinetic energy to a wheel set through a coupler and the gear box. The increase in load during train operation places higher demands on the operational life and reliability of the traction motors. Therefore, detecting the temperature of the traction motor is also very important for train driving safety. When the subway train runs at a high speed, the working condition of the gear box is bad, and according to domestic researches on the gear box, the running speed of the train is too high or too low to cause adverse effects on the internal temperature of the gear box. Under high speed drive shaft operation, particularly in hot environments, excessive gearbox temperatures can lead to seal failure, negatively affecting lubrication of bearings and gears, and reducing gearbox life.
At present, temperature measurement of key equipment at the bottom of a train is carried out in China, and most of the temperature measurement systems adopt two types of shaft temperature detection systems, namely a direct contact type shaft temperature detection system and a non-direct contact type shaft temperature detection system. The working state maintenance and calibration workload of the contact sensor under the severe environment can be very complicated, the operation cost of the subway is greatly increased, and a single contact temperature measuring point on each axle box can cause a great hot axis misjudgment risk of the temperature detection system. The non-direct contact type shaft temperature detection system mainly adopts an infrared single-point detection technology at present, and measures the temperature value of a certain point of a key component through the temperature measuring infrared rays emitted by an infrared sensor, so that the temperature measuring mode can only obtain the local temperature of a vehicle bottom component, and the heating condition of the whole subway train running part cannot be completely reflected. There are currently ways to identify the location of critical components for infrared heat map localization using traditional manual design feature extractors and ways to use deep learning neural networks, but with less accuracy.
Disclosure of Invention
The invention aims to provide a temperature detection method and a temperature detection system for a running part of a subway train based on deep learning, so that high-temperature faults of key equipment of the running part of the subway train are accurately and efficiently diagnosed.
The technical solution for realizing the purpose of the invention is as follows: a subway train running part temperature detection method based on deep learning comprises the following steps:
and 4, based on the target component extracted in the step 3, combining infrared heat map graying treatment to obtain a single-channel gray image, and calculating the temperature value of the target component.
Further, in step 1, the infrared image acquisition system includes a thermal infrared imager and a wheel sensor, the thermal infrared imager is installed at the middle position of the two rails, and the camera direction is vertically upward; the wheel sensor is arranged at a position close to the rail, and the installation direction is perpendicular to the travelling direction of the train;
the infrared thermal imager is used for acquiring infrared images of the running part of the subway train, and the wheel sensor is used for acquiring incoming signals of the subway train; when the train passes through the wheel sensor, the wheel sensor acquires a pulse signal, and the pulse signal passes through the upper computer to trigger the thermal infrared imager to start to acquire infrared images;
the infrared thermal imager is used for collecting pixel temperature points, and the temperature of a running part target component is collected in real time, so that an infrared heat map of a running part in the running process of a train is obtained, wherein the running part target component comprises an axle box, a traction motor and a gear box.
Further, in step 2, the infrared heat map of the target component is subjected to noise adding, rotation and scaling, and the processed infrared heat map is subjected to data labeling, so as to manufacture a classification label, and the specific process is as follows:
(2.1) aiming at the infrared heat map acquired by the infrared image acquisition system, carrying out processing comprising noise addition, rotation and scaling so as to realize data enhancement;
(2.2) marking data of the infrared heat map added with data by using LabelImg software, generating a xml format table file, manufacturing a classification tag motor, an axle box and a gear box, and completing the construction of a target data set;
(2.3) 3 categories were defined when using LableImg software to make classification labels, namely the categories in the voc_class.txt file; after the data is marked, the xml file with the label and the corresponding source image are stored according to the data set format of PASCAL_VOC 2007; the data folder data is built under the same-level catalog of training codes train.py, namely under the primary catalog VOCdevkit folder, and three folders are built in the data folder: annotations, imageSets and JPEGImages; the items folder stores the label frame training information, the ImageSets folder stores the training set, and the JPEGImages folder stores the test set.
Further, in step 3, an improved YOLOv5 deep learning algorithm training model is built, and the specific process is as follows:
(3.1) incorporating the attention mechanism module CBAM of the convolution module into the Backbone network of YOLOv5, inserting the attention mechanism module of the convolution module between the Backbone and the neg module of YOLOv 5;
(3.2) improving the original target boundary regression function, replacing the original GIoU loss function by adopting CIoU, wherein the GIoU loss function considers the overlapping area between the prediction frame and the corresponding ground true phase, and also considers the distance between center points and the aspect ratio of two boundary frames, and the calculation formula is as follows:
wherein IoU denotes the intersection ratio, ρ denotes the Euclidean distance between the predicted and real predicted frames, b denotes the center point of the real frame, b gt Representing the center point of the prediction frame, c representing the shortest diagonal length of the smallest frame containing ground truth and prediction frame, α being a weight parameter, v representing the similarity between the aspect ratios of the two bounding frames, w gt And h gt The width and the height of the ground real phase are represented, w and h respectively represent the width and the height of a prediction frame, and CIoU represents complete boundary regression loss;
(3.3) parameters for model training are set as follows: class = 3 for the [ yolo ] part of the yolov5.Cfg file; setting iteration times, namely setting a parameter epochs, in a training code train; saving the generated weight model.save_weights (log_dir+ 'train_weight.hs'), wherein model.save_weights represents model save weights, log_dir represents save paths, train_weight.hs represents training weights.
Further, in step 4, based on the target component extracted in step 3, a single-channel gray-scale image is obtained by combining infrared heat map gray-scale processing, and a temperature value of the target component is calculated, specifically as follows:
(4.1) carrying out gray-scale treatment on the infrared heat map to obtain a single-channel gray-scale image;
(4.2) carrying out Gaussian filtering treatment on the gray level image in the step (4.1) to improve the signal-to-noise ratio of the image;
(4.3) dividing the obtained picture in the step (4.2) into target main areas by adopting an adaptive threshold;
and (4.4) combining the gray level image in (4.1) and the pixel coordinates of the main target area segmented in (4.3), and calculating the temperature value corresponding to each pixel point.
The utility model provides a subway train running part temperature detection system based on degree of depth study, includes infrared image acquisition system, infrared heat map processing module, model construction module and temperature calculation module, wherein:
the infrared image acquisition system acquires an infrared heat map of a target part of the running part in the running process of the train;
the infrared heat map processing module is used for carrying out noise adding, rotation and scaling on the infrared heat map in the infrared image acquisition system, carrying out data marking on the processed infrared heat map, and manufacturing a classification label to obtain an infrared heat map data set;
the model building module is used for building an improved YOLOv5 deep learning algorithm training model, integrating the convolution attention module into a YOLOv5 backbone network, extracting the characteristics of a target component of a running part of a train by using CIoU as a frame regression loss function of a target, training the model according to a data set obtained by the infrared heat map processing module, and identifying the information of the target component of the running part;
and the temperature calculation module is used for calculating the temperature value of the target component based on the target component extracted from the model construction module and combining infrared heat map graying treatment to obtain a single-channel gray image.
A mobile terminal comprises a memory, a processor and a computer program stored on the memory and capable of running on the processor, wherein the processor realizes a subway train running part temperature detection method based on deep learning when executing the program.
A computer readable storage medium having stored thereon a computer program which when executed by a processor implements the steps in the deep learning based subway train running section temperature detection method.
Compared with the prior art, the invention has the remarkable advantages that: (1) Classifying and identifying key equipment based on infrared thermal images of the vehicle bottom parts, constructing and training a model by utilizing a field infrared thermal image, and integrating a CBAM attention mechanism into an original model compared with a traditional YOLOv5 algorithm to further extract the characteristics of a target and further improve the detection precision; (2) The CIoU is used as a frame regression loss function of the target, so that the prediction of the target position is more real; (3) The infrared thermal imaging system is a planar array type temperature measurement system, and can be used for multi-equipment temperature detection of a running urban rail train.
Drawings
FIG. 1 is a network structure diagram of the improved YOLOv5 algorithm of the present invention based on the attention mechanism.
Fig. 2 is a diagram showing a specific network structure of the attention mechanism of the present invention.
FIG. 3 is a graph showing the results of improved model object detection in accordance with an embodiment of the present invention.
Fig. 4 is a PR graph of the model training of the present invention.
Fig. 5 is a schematic diagram of temperature detection of a running gear device based on deep learning target detection in an embodiment of the invention.
Fig. 6 is a schematic diagram of temperature detection of a running gear apparatus according to an embodiment of the invention.
Detailed Description
The invention discloses a subway train running part temperature detection method based on deep learning, which comprises the following steps of:
and 4, based on the target component extracted in the step 3, combining infrared heat map graying treatment to obtain a single-channel gray image, and calculating the temperature value of the target component.
Further, in step 1, the infrared image acquisition system includes a thermal infrared imager and a wheel sensor, the thermal infrared imager is installed at the middle position of the two rails, and the camera direction is vertically upward; the wheel sensor is arranged at a position close to the rail, and the installation direction is perpendicular to the travelling direction of the train;
the infrared thermal imager is used for acquiring infrared images of the running part of the subway train, and the wheel sensor is used for acquiring incoming signals of the subway train; when the train passes through the wheel sensor, the wheel sensor acquires a pulse signal, and the pulse signal passes through the upper computer to trigger the thermal infrared imager to start to acquire infrared images;
the infrared thermal imager is used for collecting pixel temperature points, and the temperature of a running part target component is collected in real time, so that an infrared heat map of a running part in the running process of a train is obtained, wherein the running part target component comprises an axle box, a traction motor and a gear box.
Further, in step 2, the infrared heat map of the target component is subjected to noise adding, rotation and scaling, and the processed infrared heat map is subjected to data labeling, so as to manufacture a classification label, and the specific process is as follows:
(2.1) aiming at the infrared heat map acquired by the infrared image acquisition system, carrying out processing comprising noise addition, rotation and scaling so as to realize data enhancement;
(2.2) marking data of the infrared heat map added with data by using LabelImg software, generating a xml format table file, manufacturing a classification tag motor, an axle box and a gear box, and completing the construction of a target data set;
(2.3) 3 categories were defined when using LableImg software to make classification labels, namely the categories in the voc_class.txt file; after the data is marked, the xml file with the label and the corresponding source image are stored according to the data set format of PASCAL_VOC 2007; the data folder data is built under the same-level catalog of training codes train.py, namely under the primary catalog VOCdevkit folder, and three folders are built in the data folder: annotations, imageSets and JPEGImages; the items folder stores the label frame training information, the ImageSets folder stores the training set, and the JPEGImages folder stores the test set.
Further, in step 3, an improved YOLOv5 deep learning algorithm training model is built, and the specific process is as follows:
(3.1) incorporating the attention mechanism module CBAM of the convolution module into the Backbone network of YOLOv5, inserting the attention mechanism module of the convolution module between the Backbone and the neg module of YOLOv 5;
(3.2) improving the original target boundary regression function, replacing the original GIoU loss function by adopting CIoU, wherein the GIoU loss function considers the overlapping area between the prediction frame and the corresponding ground true phase, and also considers the distance between center points and the aspect ratio of two boundary frames, and the calculation formula is as follows:
wherein IoU denotes the intersection ratio, ρ denotes the Euclidean distance between the predicted and real predicted frames, b denotes the center point of the real frame, b gt Representing the center point of the prediction frame, c representing the shortest diagonal length of the smallest frame containing ground truth and prediction frame, α being a weight parameter, v representing the similarity between the aspect ratios of the two bounding frames, w gt And h gt The width and the height of the ground real phase are represented, w and h respectively represent the width and the height of a prediction frame, and CIoU represents complete boundary regression loss;
(3.3) parameters for model training are set as follows: class = 3 for the [ yolo ] part of the yolov5.Cfg file; setting iteration times, namely setting a parameter epochs, in a training code train; saving the generated weight model.save_weights (log_dir+ 'train_weight.hs'), wherein model.save_weights represents model save weights, log_dir represents save paths, train_weight.hs represents training weights.
Further, in step 4, based on the target component extracted in step 3, a single-channel gray-scale image is obtained by combining infrared heat map gray-scale processing, and a temperature value of the target component is calculated, specifically as follows:
(4.1) carrying out gray-scale treatment on the infrared heat map to obtain a single-channel gray-scale image;
(4.2) carrying out Gaussian filtering treatment on the gray level image in the step (4.1) to improve the signal-to-noise ratio of the image;
(4.3) dividing the obtained picture in the step (4.2) into target main areas by adopting an adaptive threshold;
and (4.4) combining the gray level image in (4.1) and the pixel coordinates of the main target area segmented in (4.3), and calculating the temperature value corresponding to each pixel point.
The invention discloses a subway train running part temperature detection system based on deep learning, which comprises an infrared image acquisition system, an infrared heat map processing module, a model building module and a temperature calculation module, wherein:
the infrared image acquisition system acquires an infrared heat map of a target part of the running part in the running process of the train;
the infrared heat map processing module is used for carrying out noise adding, rotation and scaling on the infrared heat map in the infrared image acquisition system, carrying out data marking on the processed infrared heat map, and manufacturing a classification label to obtain an infrared heat map data set;
the model building module is used for building an improved YOLOv5 deep learning algorithm training model, integrating the convolution attention module into a YOLOv5 backbone network, extracting the characteristics of a target component of a running part of a train by using CIoU as a frame regression loss function of a target, training the model according to a data set obtained by the infrared heat map processing module, and identifying the information of the target component of the running part;
and the temperature calculation module is used for calculating the temperature value of the target component based on the target component extracted from the model construction module and combining infrared heat map graying treatment to obtain a single-channel gray image.
The invention relates to a mobile terminal, which comprises a memory, a processor and a computer program stored on the memory and capable of running on the processor, wherein the processor realizes a subway train running part temperature detection method based on deep learning when executing the program.
The invention provides a computer readable storage medium, wherein a computer program is stored on the computer readable storage medium, and the program is executed by a processor to realize the steps in the subway train running part temperature detection method based on deep learning.
The present invention will be further described with reference to fig. 1 to 5 and specific examples.
Examples
The invention discloses a temperature detection method of subway train bottom equipment based on deep learning, which uses an improved YOLOv5 deep learning algorithm integrating an attention mechanism, and an improved network structure model is shown in figure 1. The detailed structure of the attention mechanism module is shown in fig. 2.
And a field image acquisition system is arranged below the running part of the train, namely at the two sides of the track at the bottom of the train and in the middle of the track. When the train runs through the acquisition system, the image acquisition system can acquire images and temperature data of the key equipment of the running part.
Step 1: an infrared image acquisition system is arranged in the track area to acquire infrared images of key components of the subway running part, including an axle box, a traction motor and a gear box.
Further, step 1 specifically operates as: the thermal infrared imager is arranged at the middle position of the two rails, and the direction of the camera is vertically upward; the wheel sensor is arranged at a position close to the rail, and the installation direction is vertical to the travelling direction of the train; the thermal infrared imager is mainly used for acquiring infrared images of the running part of the subway train, and the wheel sensor is mainly used for acquiring incoming signals of the subway train; when the train passes through the wheel sensor, the wheel sensor can obtain a pulse signal, and the pulse signal can inform the thermal infrared imager to start to collect infrared images through the upper computer.
And 2, after the infrared image is obtained, the image needs to be subjected to data labeling and label making.
Further, the specific operation of step 2 is as follows:
(1) And carrying out image preprocessing on the acquired infrared thermal image, and carrying out noise addition, rotation and scaling for image enhancement.
(2) Carrying out data annotation by using LabelImg software to generate a corresponding xml format table file; manufacturing classification labels such as a motor, a shaft box and a gear box; and finally, constructing the target data set.
(3) 3 categories were defined when using LableImg to make a class label in data calibration, namely the category in the voc_class.txt file. After the data is annotated, the tagged xml file and its corresponding source image are saved in the data set format of the paspal VOC 2007. The data folder data is established under the same-level directory of training codes train.py, namely under a first-level directory VOCdevkit folder, and three folders are established in the data folder: annotations, imageSets and JPEGImages. The items folder stores the label frame training information, the ImageSets folder stores the training set, and the JPEGImages folder stores the test set.
Step 3: the method is characterized in that the Yolov5 is improved, an algorithm of a convolution attention mechanism module is integrated, and the convolution attention module is integrated into a Yolov5 backbone network so as to improve the feature extraction capability. Meanwhile, the loss function is improved, and the CIoU is used as a target frame regression loss function.
And improving the original target boundary regression function, and replacing the original GIoU loss function by adopting CIoU. The GIoU loss function considers the distance between the center points and the aspect ratio of the two bounding boxes in addition to the overlap region between the prediction box and the corresponding ground truth. The calculation formula is as follows:
wherein IoU represents the cross-over ratio and ρ representsEuclidean distance between predicted and real predicted frames, b represents the center point of the real frame, b gt Representing the center point of the prediction frame, c representing the shortest diagonal length of the smallest frame containing ground truth and prediction frame, α being a weight parameter, v representing the similarity between the aspect ratios of the two bounding frames, w gt And h gt Representing the width and height of the ground truth, w and h representing the width and height of the prediction box, respectively, CIoU representing the complete boundary regression loss.
Step 4: and training the model and verifying the performance of the model according to the constructed training set and verification set.
Step 5: and detecting the roof image to be detected by using the best.py weight file after model training, and outputting a detection result. The relevant parameters for model training were set as follows: class = 3 for the [ yolo ] part of the yolov5.Cfg file; the iteration number can be changed in the training code train.py, namely, the parameter epochs is set; and (5) saving the generated weight: model. Save_weights (log_dir+ 'train_weight. Hs'), where model. Save_weights represents model save weights, log_dir represents save paths, train_weight. Hs represents training weights. The result of predictive recognition of the target classification is shown in fig. 3. The PR curve of the result is shown in FIG. 4.
Table 1 classification model performance comparison
Under the condition that the training time of the improved model is similar to that of a single picture, the average prediction accuracy can be improved by 1.3% in the original model, and the effectiveness of the improved model is improved.
Step 6: firstly, carrying out graying treatment on an infrared heat map to obtain a single-channel gray image; then, gaussian filtering processing is carried out on the obtained gray level image, and the signal-to-noise ratio of the image is improved; then dividing the picture subjected to Gaussian filtering treatment by adopting a self-adaptive threshold value to obtain a target main area; and finally, combining the gray level image and the pixel coordinates of the divided main area to calculate the temperature value corresponding to each pixel point. The flow chart of target temperature identification is shown in fig. 5, and the result of temperature detection identification is shown in fig. 6.
It should be appreciated that in the above description of exemplary embodiments of the invention, various features of the invention are sometimes described in the context of a single embodiment or with reference to a single figure in order to streamline the invention and aid those skilled in the art in understanding the various aspects of the invention. The present invention should not, however, be construed as including features that are essential to the patent claims in the exemplary embodiments.
Claims (8)
1. The subway train running part temperature detection method based on deep learning is characterized by comprising the following steps of:
step 1, an infrared image acquisition system of a subway train is arranged, and an infrared heat map of a target part of a running part in the running process of the train is acquired through the infrared image acquisition system;
step 2, carrying out noise adding, rotation and scaling on the infrared heat map obtained in the step 1, carrying out data marking on the processed infrared heat map, and manufacturing a classification label to obtain an infrared heat map data set;
step 3, an improved YOLOv5 deep learning algorithm training model is established, a convolution attention module is integrated into a YOLOv5 backbone network, a CIoU is used as a target frame regression loss function, characteristics of a target part of a running part of a train are extracted, the model is trained according to the data set obtained in the step 2, and the information of the target part of the running part is identified;
and 4, based on the target component extracted in the step 3, combining infrared heat map graying treatment to obtain a single-channel gray image, and calculating the temperature value of the target component.
2. The method for detecting the temperature of the running part of the subway train based on the deep learning according to claim 1, wherein in the step 1, the infrared image acquisition system comprises a thermal infrared imager and a wheel sensor, the thermal infrared imager is arranged at the middle position of two rails, and the camera direction is vertically upward; the wheel sensor is arranged at a position close to the rail, and the installation direction is perpendicular to the travelling direction of the train;
the infrared thermal imager is used for acquiring infrared images of the running part of the subway train, and the wheel sensor is used for acquiring incoming signals of the subway train; when the train passes through the wheel sensor, the wheel sensor acquires a pulse signal, and the pulse signal passes through the upper computer to trigger the thermal infrared imager to start to acquire infrared images;
the infrared thermal imager is used for collecting pixel temperature points, and the temperature of a running part target component is collected in real time, so that an infrared heat map of a running part in the running process of a train is obtained, wherein the running part target component comprises an axle box, a traction motor and a gear box.
3. The subway train running part temperature detection method based on deep learning according to claim 1, wherein in step 2, the infrared heat map of the target component is subjected to noise adding, rotation and scaling, and the processed infrared heat map is subjected to data labeling, so as to manufacture classification labels, and the specific process is as follows:
(2.1) aiming at the infrared heat map acquired by the infrared image acquisition system, carrying out processing comprising noise addition, rotation and scaling so as to realize data enhancement;
(2.2) marking data of the infrared heat map added with data by using LabelImg software, generating a xml format table file, manufacturing a classification tag motor, an axle box and a gear box, and completing the construction of a target data set;
(2.3) 3 categories were defined when using LableImg software to make classification labels, namely the categories in the voc_class.txt file; after the data is marked, the xml file with the label and the corresponding source image are stored according to the data set format of PASCALVOC 2007; the data folder data is built under the same-level catalog of training codes train.py, namely under the primary catalog VOCdevkit folder, and three folders are built in the data folder: annotations, imageSets and JPEGImages; the items folder stores the label frame training information, the ImageSets folder stores the training set, and the JPEGImages folder stores the test set.
4. The subway train running part temperature detection method based on deep learning according to claim 1, wherein in step 3, an improved YOLOv5 deep learning algorithm training model is built, and the specific process is as follows:
(3.1) incorporating the attention mechanism module CBAM of the convolution module into the Backbone network of YOLOv5, inserting the attention mechanism module of the convolution module between the Backbone and the neg module of YOLOv 5;
(3.2) improving the original target boundary regression function, replacing the original GIoU loss function by adopting CIoU, wherein the GIoU loss function considers the overlapping area between the prediction frame and the corresponding ground true phase, and also considers the distance between center points and the aspect ratio of two boundary frames, and the calculation formula is as follows:
wherein IoU denotes the intersection ratio, ρ denotes the Euclidean distance between the predicted and real predicted frames, b denotes the center point of the real frame, b gt Representing the center point of the prediction frame, c representing the shortest diagonal length of the smallest frame containing ground truth and prediction frame, α being a weight parameter, v representing the similarity between the aspect ratios of the two bounding frames, w gt And h gt The width and the height of the ground real phase are represented, w and h respectively represent the width and the height of a prediction frame, and CIoU represents complete boundary regression loss;
(3.3) parameters for model training are set as follows: class = 3 for the [ yolo ] part of the yolov5.Cfg file; setting iteration times, namely setting a parameter epochs, in a training code train; saving the generated weight model.save_weights (log_dir+ 'train_weight.hs'), wherein model.save_weights represents model save weights, log_dir represents save paths, train_weight.hs represents training weights.
5. The method for detecting the temperature of the running part of the subway train based on the deep learning according to claim 1, wherein in the step 4, based on the target component extracted in the step 3, a single-channel gray image is obtained by combining infrared heat map gray processing, and the temperature value of the target component is calculated as follows:
(4.1) carrying out gray-scale treatment on the infrared heat map to obtain a single-channel gray-scale image;
(4.2) carrying out Gaussian filtering treatment on the gray level image in the step (4.1) to improve the signal-to-noise ratio of the image;
(4.3) dividing the obtained picture in the step (4.2) into target main areas by adopting an adaptive threshold;
and (4.4) combining the gray level image in (4.1) and the pixel coordinates of the main target area segmented in (4.3), and calculating the temperature value corresponding to each pixel point.
6. The subway train running part temperature detection system based on deep learning is characterized by comprising an infrared image acquisition system, an infrared heat map processing module, a model building module and a temperature calculation module, wherein:
the infrared image acquisition system acquires an infrared heat map of a target part of the running part in the running process of the train;
the infrared heat map processing module is used for carrying out noise adding, rotation and scaling on the infrared heat map in the infrared image acquisition system, carrying out data marking on the processed infrared heat map, and manufacturing a classification label to obtain an infrared heat map data set;
the model building module is used for building an improved YOLOv5 deep learning algorithm training model, integrating the convolution attention module into a YOLOv5 backbone network, extracting the characteristics of a target component of a running part of a train by using CIoU as a frame regression loss function of a target, training the model according to a data set obtained by the infrared heat map processing module, and identifying the information of the target component of the running part;
and the temperature calculation module is used for calculating the temperature value of the target component based on the target component extracted from the model construction module and combining infrared heat map graying treatment to obtain a single-channel gray image.
7. A mobile terminal comprising a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that the processor implements the deep learning based subway train running section temperature detection method according to any one of claims 1 to 5 when executing the program.
8. A computer-readable storage medium, on which a computer program is stored, characterized in that the program, when executed by a processor, implements the steps in the deep learning-based subway train running section temperature detection method according to any one of claims 1 to 5.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310394440.9A CN116245875A (en) | 2023-04-13 | 2023-04-13 | Subway train running part temperature detection method and system based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310394440.9A CN116245875A (en) | 2023-04-13 | 2023-04-13 | Subway train running part temperature detection method and system based on deep learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116245875A true CN116245875A (en) | 2023-06-09 |
Family
ID=86635254
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310394440.9A Pending CN116245875A (en) | 2023-04-13 | 2023-04-13 | Subway train running part temperature detection method and system based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116245875A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118247572A (en) * | 2024-04-11 | 2024-06-25 | 阿谱斯(上海)通信技术有限公司 | Intelligent digital keyboard pressing system and method based on AI deep learning |
-
2023
- 2023-04-13 CN CN202310394440.9A patent/CN116245875A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118247572A (en) * | 2024-04-11 | 2024-06-25 | 阿谱斯(上海)通信技术有限公司 | Intelligent digital keyboard pressing system and method based on AI deep learning |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Wu et al. | Hybrid deep learning architecture for rail surface segmentation and surface defect detection | |
| CN109658387B (en) | Method for detecting defects of pantograph carbon slide plate of electric train | |
| EP3964908A1 (en) | Deep hybrid convolutional neural network for fault diagnosis of wind turbine gearboxes | |
| CN202793389U (en) | On-line routing inspection device for high-speed rail overhead line system | |
| CN111260629A (en) | Pantograph structure abnormity detection algorithm based on image processing | |
| Luo et al. | An ultrasmall bolt defect detection method for transmission line inspection | |
| CN116245875A (en) | Subway train running part temperature detection method and system based on deep learning | |
| Li et al. | Joint attention feature transfer network for gearbox fault diagnosis with imbalanced data | |
| CN110271929A (en) | A kind of city elevator maintenance QA system based on big data driving | |
| CN114529821A (en) | Offshore wind power safety monitoring and early warning method based on machine vision | |
| CN114429504A (en) | Trace leakage diagnosis method for hydrogen system of fuel cell automobile | |
| CN111738089A (en) | Subway inspection AGV fault detection method | |
| CN113530850B (en) | Centrifugal pump fault diagnosis method based on ESA and stacked capsule self-encoder | |
| Wang et al. | Severe rail wear detection with rail running band images | |
| CN111565024A (en) | Solar cell panel fault detection and operation efficiency prediction device and method | |
| CN111597948A (en) | Fault detection and classification method for bearing vibration signals | |
| Charitha et al. | Machine learning based fault detection in induction motor using thermal imaging | |
| CN117150340A (en) | Method and device for diagnosing faults of small samples of switch machine | |
| Li et al. | Pantograph slide plate abrasion detection based on deep learning network | |
| CN108198164A (en) | A kind of detection device and method of cable tunnel cable integrity | |
| Wang et al. | Image recognition of wind turbines blade surface defects based on mask-RCNN | |
| CN115201213A (en) | Passenger ropeway steel wire rope detection device and method | |
| CN113674329A (en) | Vehicle driving behavior detection method and system | |
| CN112861620A (en) | System and method for identifying damage of road well cover | |
| Wenqiang et al. | An Automatic Fracture Defect Detection Approach for Current-carrying Rings of Catenary Droppers Using A Multi-task Neural Network |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |