Unsupervised defect detection method

Abstract

The invention discloses an unsupervised defect detection method, which comprises the following steps of: acquiring a real-time track picture, and intercepting the real-time track picture by using a fastener detection algorithm to obtain a fastener picture; a full connection layer in the convolutional Encoder-Decoder structure is replaced by a convolutional layer, and the fastener picture is reconstructed by using the optimized convolutional Encoder-Decoder structure, so that a reconstructed picture is obtained; constructing an SSIM similarity calculation formula, calculating the similarity between the reconstructed picture and the fastener picture, judging whether the fastener picture has defects or not according to a preset threshold range, if yes, marking the fastener picture as having defects, and if not, marking the fastener picture as not having defects; according to the invention, on the basis of a single end-to-end network, the input of an original image and the detection output of an object category are completed, the detection precision is greatly improved, and the detection accuracy is improved by applying a convolutional Encoder-Decoder structure and a fastener detection algorithm. The problem that the false drop rate and the omission rate are high in the process of performing model training through manually marking defects by singly using the convolutional neural network in the prior art is solved.

Description

technical field [0001] The invention relates to the field of traffic track detection, in particular to an unsupervised defect detection method. Background technique [0002] In the prior art, in the process of developing a subway track defect detection algorithm and using the existing single-use convolutional neural network to manually mark defects for model training, the objective conditions lead to few defect samples, which is difficult to collect and eventually leads to model generalization. However, in practical applications, there are unsupervised defect detection methods with high false detection rate and high missed detection rate. SUMMARY OF THE INVENTION [0003] In view of the above deficiencies in the prior art, the present invention provides an unsupervised defect detection method, which completes from the input of the original image to the output of the position and category of the object through a single end-to-end network, and provides a high-precision defec...

AI Technical Summary

Problems solved by technology

[0003] Aiming at the above-mentioned deficiencies in the prior art, the present invention provides an unsupervised defect detection method, which completes the input from the original image to the output of the object position and category through a separate end-to-end network, providing a high-quality The unsupervised defect detection method with high detection accuracy solves the problem that existing defect detection algorithms must use defective samples for training: few training samples and low generalization

Images