Research On Railway Safety Monitoring Based On Deep Learning

Railway is an important part of Chinese transportation system,so the safety detection of railway system is particularly important.The existing rail transit safety monitoring system^[1,2]is only suitable for acoustic event recognition in the environment of high SNR,but in low SNR and non-standard optical cable laying it is low recognition rate.The deep learning algorithm model cannot be monitored in real time because of the large amount of data.This thesis includes:First,to solve the problem of large amount data,I used the digital signal processing to detect speech and non-speech signal,before audio data was inputed into deep learning model.Then the data which input into deep learning mode were greatly reduced.Secondly,to make the railway monitoring system in the complex environment has a good performance,I designed the Intersection-over-complement Convolutional Neural Network algorithm model.After preprocessing the audio signal,I extracted the Cepstrum coefficient of Mayer frequency and the gray scale of five consecutive acquisition nodes,and took these two features as the input of IOC-CNN model,so as to carry out real-time safety monitoring of the railway system.