基于改进DeepLabv3+算法的起重机锈迹检测
赵章焰 1王成豪1
作者信息
- 1. 武汉理工大学交通与物流工程学院 武汉 430063
- 折叠
摘要
室外工作的起重机金属结构易产生锈蚀现象,严重的锈蚀会导致结构承载能力显著降低,从而引发灾难性事故.文中针对当前起重机人工锈迹巡检中存在的漏检、误检和费时等问题,提出一种基于改进DeepLabv3+算法的自动化锈迹检测方法.该方法依托于机器视觉,将原始DeepLabv3+的骨干网络替换为幽灵网络(GhostNet)以提升网络的轻量化程度;使用特征金字塔网络(FPN)进行特征提取,用于抑制噪声和背景对锈迹提取的不良干扰;引入空间感知独立自注意机制(SSA)来提高网络区域感知性能;最后使用特征融合(Add)代替原始网络的特征堆叠来降低算法参数量.将所提方法应用于室外起重机锈迹检测,结果表明所提算法的检测性能优于原始算法和其他经典语义分割算法,具有重要的工程应用价值.
Abstract
The metal structure of cranes working outdoors is prone to corrosion,and serious corrosion will significantly reduce the bearing capacity of the structure,which will lead to catastrophic accidents.In this paper,an automatic rust detection method based on improved DeepLabv3+algorithm is proposed,considering the problems of missing detection,false detection and time-consuming in manual rust inspection of cranes.Based on machine vision,this method replaces the original backbone network of DeepLabv3+with GhostNet network to improve the lightweight of the network.Feature Pyramid Networks(FPN)were used for feature extraction,which is used to suppress the adverse interference of noise and background on rust extraction.Stand-alone self-attention(SSA)was introduced to improve the performance of network area awareness.Finally,feature fusion(Add)was used to replace the feature stack of the original network to reduce the parameters of the algorithm.The proposed method was applied to outdoor crane rust detection,and the results show that compared with the original algorithm and other classical semantic segmentation algorithms,the proposed algorithm has better detection performance and has important engineering application value.
关键词
起重机/锈迹检测/改进的DeepLabv3+/幽灵网络/特征金字塔网络Key words
crane/rust detection/improved DeepLabv 3+/GhostNet/characteristic pyramid network引用本文复制引用
基金项目
省部级基金()
中交集团首个揭榜挂帅科技攻关项目(2021-ZJKJ-JBGS01)
出版年
2024
国家科技期刊平台
NSTL
万方数据