Tunnel surface defect detection method based on convolutional neural network

Abstract

The invention discloses a tunnel surface defect detection method based on a convolutional neural network. The method comprises the steps of obtaining tunnel image data; preprocessing the acquired image data, and dividing a data set into a test data set and a training and verification data set; and building a full convolution defect detection network, carrying out network model training, inputtingthe tunnel surface image into a convolution neural network, and outputting a predicted category and a defect area. According to the method, whether the tunnel surface image has defects or not can be quickly judged, the defect types and the defect positions are provided, automatic analysis of the defects in the tunnel surface image is completed, the labor cost is saved, interference of subjective factors is eliminated, and the method has the advantages of being high in efficiency, accuracy and practicability.

Description

technical field [0001] The invention relates to a tunnel surface defect detection method, in particular to a tunnel surface defect detection method based on a convolutional neural network, which belongs to the field of image processing and analysis. Background technique [0002] Tunnel is an important railway facility, and its condition directly affects railway traffic safety and transportation efficiency. With the implementation of my country's railway speed-up strategy, higher requirements have been put forward for the safety of trains. At the same time, the increase in operating speed and the operation of heavy-duty trains have increased the damage to the track, leading to aggravated deterioration of the track state. Therefore, regular inspection of tunnels, early detection of damage and timely repairs to avoid accidents have become a basic task in railway work. [0003] At present, the technical methods of tunnel surface defect detection mainly include artificial visual...

AI Technical Summary

Problems solved by technology

The visual method is labor-intensive, dangerous, time-consuming, low-efficiency, and the measurement results are subject to subjective influence; the magnetic particle method has high operating costs, and at the same time, the classification accuracy is low and the detection speed is low; the eddy current method has high-frequency excitation signal, which makes the system structure and signal processing more complex, and the detection efficiency is relatively low

Images