Research On Train Wheel Tread Damage Detection Algorithm Based On Deep Learnin

With the diversified development of railway technology,there are more and more types of rail transit,which are not only widely used in long-distance land transportation,but also in medium and short distance urban public transportation.As an important support and running component of rail transit,wheel sets may experience high-speed rolling contact with the track in complex environments for a long time,which may cause damage such as tread scratches,scratches,and peeling,affecting operational safety.Therefore,achieving rapid and accurate detection of tread damage is particularly important for ensuring the safe operation of locomotives.Traditional detection of damage to train wheel tread has problems such as low accuracy and slow efficiency.In response to the limitations of existing technologies,this thesis will use deep learning methods to detect and identify damage to train wheel tread.The main work includes the following aspects:(1)Establishment of a dataset for wheel tread damage detection.In order to meet the training requirements of deep learning algorithms and obtain richer tread damage images,this thesis uses an improved convolutional generative adversarial network to enhance the tread damage dataset based on tread images containing real damage.In order to achieve a more similar feature distribution between the generated virtual image and the real image,a self attention mechanism is introduced to enhance the modeling ability between the generator and discriminator in the separated regions;In order to improve the training stability of the network,Wasserstein distance is used as a new loss function.Experimental results show that the generative network can generate tread images with damaged texture features.After obtaining the damaged target image,the mask image of the target is converted into a Boolean value to remove the background of the target image and complete coverage fusion with the normal tread image.(2)Wheel set tread damage detection based on YOLOv7.In order to improve the accuracy of various types of tread damage detection,the Sim AM attention mechanism is first introduced,while paying attention to the importance of each channel and spatial position feature,so that important target features occupy a greater proportion in network processing,thereby enhancing the network’s feature learning ability for the target area;Then the loss function of the border regression is changed to Focal EIo U,and more border information is fused to improve the accuracy of prediction;Finally,a dynamic head for object detection based on attention mechanism is added to dynamically and adaptively adjust attention weights,making the model more flexible and accurate while reducing the number of parameters.The experiment shows that the average accuracy m AP value of the improved YOLOv7 on the test set reaches 94.6%,and the detection frame rate FPS is 43.5.(3)Design of a system for detecting wheel tread damage.In order to facilitate and efficiently complete the classification and identification of wheel tread damage,Pyqt5 is used to write a real-time detection UI interface.Firstly,extract the tread area from the image data collected by the camera,and then input the preprocessed image into the trained network model to complete the detection of tread damage.