A TWO-PCSN-STFF Algorithm for Steel Industry Yard Vehicle Action Recognition
作者单位：College of Control Science and Engineering Zhejiang University
会议名称：《第35届中国过程控制会议》
会议日期：1000年
学科分类：08[工学] 0806[工学-冶金工程] 080203[工学-机械设计及理论] 080601[工学-冶金物理化学] 0802[工学-机械工程]
关 键 词：Vehicle Action Recognition Spatiotemporal Feature Fusion Model Lightweight Attention Mechanism
摘 要：At present, iron and steel enterprises usually rely on manual observation to monitor the dust generated by largescale vehicle activities in the material store yard. However, this method is labor-intensive and cannot provide uninterrupted24-hour monitoring. Therefore, there is an urgent need for an intelligent monitoring system to identify the actions of vehicles on industrial sites, enabling uninterrupted unmanned monitoring and alert systems for dust in the yard. This paper proposes a novel two-stream vehicle action recognition model, incorporating a pruning-channel-separated convolution(PCSN) mechanism and a spatiotemporal feature fusion module(STFF), to process real-time video streams from steel yards. The proposed model consists of two specifically designed paths: the environment path, which extracts spatial features, and the action path, which extracts temporal features. The STFF is then employed to integrate the temporal characteristics of the action path into the environment path, thereby facilitating efficient fusion of spatiotemporal features. Additionally, the PCSN mechanism is designed to optimize the original Resnet 3D module during convolution operations. Ultimately, while maintaining the method s accuracy,the proposed approach is well-suited for deployment and real-time monitoring, providing valuable information for the subsequent dust suppression treatment.