1. A BiFPN-SECA detection network for foreign objects on top of railway freight vehicles.
- Author
-
Liu, Sheng, Yang, Yiqing, Cao, Ting, and Zhu, Yi
- Abstract
Foreign object detection on top of railway freight vehicles is critical to ensuring the safety and efficiency railway transportation. However, this task is mainly carried out manually, facing major challenges such as low accuracy and false positives and false negatives. To address the above problems, this paper proposes a novel BiFPN-SECA Network, Bidirectional Feature Pyramid Network with Squeeze-and-Excitation Channel Attention, which integrates an attention mechanism combining contextual features and coordinate information. This innovative approach improves detection performance by effectively capturing and emphasizing key features. The ordinary convolutional layers in the coordinate attention mechanism are improved in SEnet to increase the attention to the regions of interest. Second, an improved coordinate attention mechanism module is embedded before the BiFPN sampling layer and after the backbone output layer to enhance key feature extraction. In addition, a dataset of foreign objects in railway freight vehicles was established, including bag, stone and torn. Experimental results verify that the BiFPN-SECA Network is able to improve detection accuracy, with an accuracy of 94.57 % , demonstrating its potential to significantly enhance the safety and reliability of railway freight. [ABSTRACT FROM AUTHOR]
- Published
- 2024
- Full Text
- View/download PDF