Research On Intelligent Recognition Algorithm Of Railway Images Based On Convolutional Neural Network

In recent years,with the rapid development of China's railway system,the railway mileage and the coverage of the railway network have increased significantly.Moreover,due to the demands of the public on the degree of environmental protection,operation speed,and transportation capability of railway systems,the proportion of electrified railways in all railways in operation is increasing.Under this background,how to improve the safety of railway system operation,the degree of automation of detection,the accuracy of foreign object invasion and equipment fault detection,the timeliness of alarm have become the current research hotspots.The traditional way of relying on staff for manpower detection has the disadvantages of low efficiency,high cost,long detection cycle,and so on,which is far from meeting the requirements of today's railway systems.Therefore,the automatic detection method relying on the railway system image for intelligent identification attracts much attention.After learning the principle of convolutional neural network,this thesis believes that if the convolutional neural network is applied to the intelligent identification algorithm of railway images,it will greatly improve the recognition accuracy of the original detection method,and on this basis Research and experimentation.The main work of this article is as follows:(1)Introduced the basic concept and structure of convolutional neural network,and elaborated the function of each part.The advantages of convolutional neural networks in object feature learning and object recognition are analyzed.It is clear that the convolutional neural network is used as the main technical means for this study.(2)Aiming at the task of foreign body intrusion detection in railway system,this thesis proposes to distinguish the video background first,apply the moving target detection algorithm in the dangerous area of the video image for video preprocessing,and then identify the need for alarm through the convolutional neural network to identify the object.Then compares several moving target detection algorithms,and compares the accuracy of the detection results,the algorithm time-consuming and so on.(3)Based on the theory of convolutional neural network,adopting the design concept of the excellent model,combining the engineering complexity,practical requirements,and hardware configuration,the design of each component of the convolutional neural network and the construction of the entire network.According to the specific problems,the loss function,classifier,training method and optimization algorithm of convolutional neural network are determined.After training the parameters,the test of the image test library and the live video is completed,which verifies the feasibility of the foreign object intrusion intelligent recognition algorithm proposed in this thesis.(4)In combination with the actual engineering problems,an intelligent identification algorithm for contact network equipment faults based on convolutional neural network is designed.The on-site insulator and hanging string pictures were used for labeling,network parameters were trained,and the sample overcharging method was used to avoid overfitting of the network due to uneven sample size.Finally,the experiment verifies that the recognition rate of the algorithm to the insulator and hanging string fault reaches 90.2%and 96%respectively.Satisfy the actual engineering requirements,and also verify that the convolutional neural network can improve the efficiency and intelligence of the railway system detection.