Method and device for generating binaural pin defect sample based on generative adversarial network

Abstract

The invention discloses a method and device for generating a binaural pin defect sample based on a generative adversarial network, and the method comprises the steps: collecting a binaural pin sample image, and obtaining a binaural pin sample image data set, wherein the binaural pin sample image data set comprises a normal sample image data set and a defect sample image data set; training a CycleGAN model by using the binaural pin sample image data set to obtain a paired sample image data set, wherein the paired sample images comprise a normal sample image and a correspondingly generated defect sample image; and training a Pix2Pix network model by using the paired sample image data set to obtain a binaural pin defect sample based on the generative adversarial network. According to the method, characteristics of various generative adversarial network models are combined, defect sample images with vivid corresponding effects can be effectively generated by processing normal samples, a balanced data set with rich characteristics is provided for subsequent training of a defect detection model, and thus the model obtains a good training effect.

Description

technical field [0001] The present invention relates to the technical field of image processing, in particular to a method and device for generating a double ear pin defect sample based on a generative adversarial network. Background technique [0002] This section is intended to provide a background or context to embodiments of the invention that are recited in the claims. The descriptions herein are not admitted to be prior art by inclusion in this section. [0003] Rotating double lugs are important fasteners in catenary suspensions for high-speed railways. Due to the vibration and impact during the long-term operation of the train, catenary components may loosen and fall off, which brings great hidden dangers to the safe operation of the train. It is particularly important to find defects in time and take elimination measures. Traditional railway track equipment maintenance relies on manual inspection, which is difficult to adapt to the trend of high-speed, accurate an...

AI Technical Summary

Problems solved by technology

Due to the vibration and impact during the long-term operation of the train, it may cause the catenary components to loosen and fall off, which will bring great hidden dangers to the safe operation of the train. It is particularly important to find defects in time and take troubleshooting measures

Traditional railway track equipment maintenance relies on manual inspection, which is difficult to adapt to the trend of high-speed, accurate and automated inspection

In recent years, many defect detection methods based on deep learning have been proposed and put into practical use, but these methods often have high data requirements, and in the actual production environment, the number of real defect samples is less than that of normal samples. When the set is seriously unbalanced, it is difficult for the deep learning model to achieve good training results.

[0004] Traditional data enhancement methods, such as random cropping, rotation, flipping, and local deformation, can only simply increase the number of samples, but cannot increase the diversity of defect sample features

In addition to data enhancement, there is also a method of directly adding defect information to the original image by modeling defect information, extracting qualitative visual features, but this type of method is generally only suitable for defect features with relatively simple shapes, and the generated image is unnatural. Difficult to give a visually pleasing result

Images