Classification of Trackside Equipment Based on Convolutional Neural Network
Weidong Li, Jinshuang Li, Yang Liu · 2020
Aiming at the problems of existing trackside equipment intelligent recognition and classification, such as the problem of less track recognition and classification, a method for classifying trackside equipment based on deep convolutional neural networks is proposed. First, image enhancement transformation is performed on the original image to improve data utilization, expand the training set, and preprocess the sample images. Secondly, with the help of deep learning, the convolutional neural network is designed and improved to obtain a convolutional neural network composed of two convolutional layers, two pooling layers, and a fully connected layer. Finally, training is performed to obtain accuracy execution results and training models, and test sample pictures are used to make predictions to obtain classification results. According to the classification results of four trackside equipments such as turnouts, catenary, signal lights, and switch machines, it shows that the method has high recognition accuracy and strong robustness, thus verifying the effectiveness and practicability of the method. The effects of different network layers and the size of the convolution kernel on the performance and accuracy of the convolutional neural network were studied by experimental simulations. The experimental results show that the 5-layer network structure and the 9 h 9 size convolution kernel have the highest classification accuracy.