Research On Image Defect Detection Technology Of Railway Track Surface

The railway is an important cornerstone of the development of the national economy,and timely maintenance of rail defects is an important part of ensuring the safety of railway transportation.With the rapid prosperity of China's economy,high-speed,heavy-duty,and dense road networks have increasingly become the direction of future railway development and construction.At present,the traditional inefficient and complex inspection methods are difficult to meet the needs of rail inspection.The deep learning methods that have emerged in the field of target recognition in recent years have achieved leapfrogging in detection results.In this paper,high-speed linear array camera technology is used to quickly acquire the rail image,and a deep convolution generation adversarial network is used to enhance the rail defect samples.The candidate region-based deep convolutional neural network(Faster Region-Convolutional Neural Networks,Faster R-CNN)The algorithm completes the identification of the surface defect area of the rail,and completes an efficient image-type railway track surface defect detection system.First of all,on the premise of learning from the previous group's experience with the area array camera-based rail image acquisition device,it switched to a high-speed linear array camera and used an external trigger device to build an image acquisition system.Aiming at the problem of large differences in the quality of the images collected by the current rail inspection vehicles,the light effect is mainly considered,and the quality of the collected images is improved by adding a hood and an adjustable light source supplementary device.Under the condition of line frequency triggering,the linear array camera can realize the high-speed acquisition of the rail image under the premise of ensuring the image quality.Secondly,the YOLO,SSD,and R-CNN series network architectures are introduced,the advantages and disadvantages of the above methods are analyzed and summarized,and the regional candidate-based deep convolutional network architecture Faster R-CNN is selected as the deep learning network architecture in this paper.The model is described in detail.In addition,in view of the problem of scarce sample defects caused by the timely maintenance of rail defects,according to the different characteristics of fasteners and rail surface defects,the gray surface statistical method is used to extract the rail surface area.The results were corrected.The rotation defect,Gaussian transform,and noise are used to expand the fastener defect samples;the improved deep convolution generative adversarial network(DCGAN)is used to enhance the track surface defect samples.Finally,according to the complex texture features of the fastener defect features,a deep residual network ResNet50 network was used to extract the fastener defect features,and the model training was completed under the Faster R-CNN framework.According to the relatively single defect feature of rail surface,the traditional VGG16 network is used to extract the defect feature.The experimental results show that the detection accuracy of the rail inspection system in this paper has been greatly improved compared with the traditional method,and has a strong generalization,which reduces the workload of the rail inspection operator on the premise of ensuring the effect of the rail inspection.