Intelligent Information Processing System for Monitoring Rail Tracks

The article deals with the operational processing of visual information during the inspection of a railway track. The structure of an intelligent information system for monitoring rail fasteners is proposed. The system is installed in the detector car and contains an automated workstation, a diagnostic neural network, and four video cameras for receiving visual information. The problem of operational recognition of rail fastening system defects is solved by using a deep convolutional network. The procedure for preparing training and test datasets is described. The problem of a small number of defective rail fastener images was solved using the technique of synthesizing augmented images. The experiments with the intellectual information system and artificial neural network were conducted. Determination of the classification accuracy when system working in real-time at high speeds of the detector car has been carried out. A confusion matrix was used to assess the quality of the rail fastener classification.