Research and application of unsupervised learning algorithm for high-speed train outward appearance foreign object detection

Research and application of unsupervised learning algorithm for high-speed train outward appearance foreign object detection

연구 분야: Artificial Intelligence

학회: AICI '25: Proceedings of the 2025 International Conference on Artificial Intelligence and Computational Intelligence

In view of the randomness and difficulty in predicting the detection of outward appearance foreign objects in high-speed trains, this paper designs a feature storage detection method based on an unsupervised learning detection algorithm. This method used the embedded feature extraction network to extract low-dimensional and high-dimensional information from the shallow network and the deep network respectively. After calculation, an averaged feature distribution is obtained, and this distribution represents the comprehensive features of all training positive samples. After comparing with the image to be inspected, the location information of possible outward appearance objects could be obtained. Simulation verification and real scene testing show that the detection accuracy of this algorithm is good and meets the application requirements.