Memory Mechanisms Based Few-Shot Continual Learning Railway Obstacle Detection

Efficient and accurate railway track obstacle detection is crucial for ensuring travel safety. In order to solve the problem of small amount of track obstacle data and variable types of obstacles, we introduce the concept of few-shot continual learning into the object detection algorithm. This paper introduces a novel memory mechanisms-based multi-Domain few-shot continual learning algorithm. By simulating working memory and long-term memory processes, our method combines few-shot learning with continual adaptation to varying obstacles. The proposed double-loop approach reduces catastrophic forgetting, enhances model generalization, and improves detection accuracy. This innovation holds promise for advancing railway safety through effective obstacle detection. The proposed method showcases remarkable efficiency and accuracy in the detection of railway obstacles. Our experimental results demonstrate significant improvements compared to other methods.