Research On Image Recognition Of Wheelset Tread Damage Of Locomotive.Based On Deep Learning

As an important support and running part of the locomotive,the locomotive wheel has been in high-speed rolling contact with the track for a long time in a complex environment,and its tread damage has become one of the important factors affecting the safety and comfort of the locomotive.Existing manual auxiliary instrument detection methods have the problems of low detection efficiency and high labor intensity.To this end,this paper carried out research on image recognition of locomotive wheel tread damage based on deep learning.The specific research contents are as follows:(1)Aiming at the problem that wheelset tread damage is difficult to locate and there may be multiple damage areas,a Canny-YOLOv3 based tread damage image detection algorithm is proposed.This method mainly extracts candidate frames from the wheelset tread damage area.First,the Canny edge detection algorithm is used to extract and segment the tread area.Further,according to the wheel set tread image extracted by the Canny algorithm,the YOLOv3 target detection algorithm is proposed to extract the candidate frame for the damaged area on the tread surface,so as to realize the wheel set tread Accurate positioning and segmentation of damaged areas.Experimental results show that the algorithm can effectively extract multiple damage areas on the tread,and also has a strong ability to detect small damage.(2)Aiming at the problems of misselection of the tread damage candidate frame in the previous section and difficulty in identifying the damage category,an improved WGAN and Dense Net tread image damage recognition framework is proposed.First,create a new damage data set for the target image selected by the frame in YOLOv3,and use the WGAN-GP network to enhance the damage data set to increase the diversity of image samples;The set performs three-category image recognition,which is divided into three types: pseudo damage,abrasion and stripping.Experiments show that the framework can generate images that are highly similar to the original damage,and can eventually identify tread damage types well.(3)According to the image detection and recognition model of locomotive wheel tread damage proposed in this paper,a prototype system for wheel set tread damage detection was designed.The system uses the Microsoft.NET Winform interface design framework for user interface design,and the background logic design uses C # and Python language,so that the models and algorithms mentioned in this article are packaged in the prototype system in the form of modules,and verified by experiments and application tests.The effectiveness and practicality of the system.