基于强语义特征方法的大型港口机械故障检测算法  

Fault detection algorithm of large port machinery based on strong semantic feature method

作　　者：孙飞越 袁新杰 黄雨晴 祝锋[1] Sun Feiyue;Yuan Xinjie;Huang Yuqing;Zhu Feng

机构地区：[1]武汉理工大学,武汉430063

出　　处：《起重运输机械》2024年第17期63-68,共6页Hoisting and Conveying Machinery

摘　　要：文中提出了一种基于强语义特征融合的分割方法,用于大型港口机械故障的视觉化检测。该方法在DeepLabV3+架构的基础上进行了创新性的改进,包括双尺度特征提取和特征融合注意力机制(FFM),通过使用自采集构建的广州港大型起重机故障数据集,成功检测了港口机械的锈蚀、油漆脱落和裂缝故障,展现了高准确率和鲁棒性。实验结果表明:该方法相比原始DeepLabV3+架构展现了更高的准确率和泛化能力,为大型港口机械故障的视觉化检测提供了一种有效的深度学习解决方案,并为未来此领域的研究提供了新的思路和方向。In this paper,a segmentation method based on strong semantic feature fusion is proposed for visual detection of mechanical faults in large ports.Based on DEPLABV 3+architecture,this method is innovatively improved,including dual-scale feature extraction and feature fusion attention mechanism(FFM).By using the self-collected fault data set of Guangzhou Port’s large crane,the port machinery faults such as rust,paint peeling and cracks were successfully detected,showing high accuracy and robustness.The experimental results show that this method has higher accuracy and generalization ability compared with the original DeepLABV 3+architecture,which provides an effective deep learning solution for visual detection of mechanical faults in large ports,and provides new ideas and directions for future research in this field.

关 键 词：港口机械 故障检测 强语义特征融合 深度学习 视觉化检测 

分 类 号：U653.921[交通运输工程—港口、海岸及近海工程] TH165+.3[交通运输工程—船舶与海洋工程]