A lightweight YOLOX-based model for detection of metallic gaskets on High-speed EMU wheelset

Abstract In order to solve the requirements of real-time detection on metallic gaskets wheelset assembly under condition limited hardware resources, here we propose an improved lightweight YOLOX-like network that can be deployed mobile devices with low computing resource consumption. Firstly, basic components ShuffleNetV2 structure are trimmed reduce computation intensity and number training parameters. Thus is used as backbone YOLOX. The prediction layer for $$20\times 20$$ 20 × size large targets also removed from head this further calculation time improve inference speed. Besides, channel-attention enhancement module efficient channel attention added into capability feature extraction accuracy target detection. Finally, verification proposed model carried out terminal devices.The results show algorithm in paper not only ensures accuracy, but greatly reduces parameters it particularly run rapidly low-cost computation.