Supervised Anomaly Detection Method Combining Generative Adversarial Networks and Three-Dimensional Data in Vehicle Inspections
Authors:
Yohei Baba,
Takuro Hoshi,
Ryosuke Mori,
Gaurang Gavai
Abstract:
The external visual inspections of rolling stock's underfloor equipment are currently being performed via human visual inspection. In this study, we attempt to partly automate visual inspection by investigating anomaly inspection algorithms that use image processing technology. As the railroad maintenance studies tend to have little anomaly data, unsupervised learning methods are usually preferred…
▽ More
The external visual inspections of rolling stock's underfloor equipment are currently being performed via human visual inspection. In this study, we attempt to partly automate visual inspection by investigating anomaly inspection algorithms that use image processing technology. As the railroad maintenance studies tend to have little anomaly data, unsupervised learning methods are usually preferred for anomaly detection; however, training cost and accuracy is still a challenge. Additionally, a researcher created anomalous images from normal images by adding noise, etc., but the anomalous targeted in this study is the rotation of pi** cocks that was difficult to create using noise. Therefore, in this study, we propose a new method that uses style conversion via generative adversarial networks on three-dimensional computer graphics and imitates anomaly images to apply anomaly detection based on supervised learning. The geometry-consistent style conversion model was used to convert the image, and because of this the color and texture of the image were successfully made to imitate the real image while maintaining the anomalous shape. Using the generated anomaly images as supervised data, the anomaly detection model can be easily trained without complex adjustments and successfully detects anomalies.
△ Less
Submitted 22 December, 2022;
originally announced December 2022.
Railway Anomaly detection model using synthetic defect images generated by CycleGAN
Authors:
Takuro Hoshi,
Yohei Baba,
Gaurang Gavai
Abstract:
Although training data is essential for machine learning, railway companies are facing difficulties in gathering adequate images of defective equipment due to their proactive replacement of would be defective equipment. Nevertheless, proactive replacement is indispensable for safe and undisturbed operation of public transport. In this research, we have developed a model using CycleGAN to generate…
▽ More
Although training data is essential for machine learning, railway companies are facing difficulties in gathering adequate images of defective equipment due to their proactive replacement of would be defective equipment. Nevertheless, proactive replacement is indispensable for safe and undisturbed operation of public transport. In this research, we have developed a model using CycleGAN to generate artificial images of defective equipment instead of real images. By adopting these generated images as training data, we verified that these images are indistinguishable from real images and they play a vital role in enhancing the accuracy of the defect detection models.
△ Less
Submitted 24 February, 2021;
originally announced February 2021.
