首页|基于多通道轴向注意力的钢索表面损伤视觉检测方法

基于多通道轴向注意力的钢索表面损伤视觉检测方法

扫码查看
工业场景中,基于视觉图像的钢索表面损伤实时检测预警存在许多挑战,如特征提取难度大、存在漏检误检、难以满足实时性要求等.为此,提出了一种基于多通道轴向注意力的钢索损伤视觉检测方法.首先,根据钢索的形态学特点提出了一种多通道轴向注意力机制,以更高效地提取和关注更加关键的钢索轴向区域特征信息;然后,通过中心化特征金字塔模块进行多尺度融合,提取不同尺度的特征信息;最后,将轻量化的MobileNetV3特征提取网络与无锚框检测头相结合,兼顾了检测速度与精度.在钢索损伤可见光图像数据集上开展的实验结果表明:该方法相较于常用目标检测的最佳算法提高了 1.99%的准确率,能够更好地检测钢索表面的损伤.
Visual detection of surface damage on wire ropes based on multi-channel axial attention
In industrial scenarios,there were many challenges,such as difficulty of feature extraction,existence of leakage or misdetection and failure to meet the real-time requirements,in real-time detec-tion of wire ropes surface damage early warning based on visual images.To solve these problems,a visual detection algorithm for wire rope damage was proposed based on multi-channel axial attention.Firstly,on the basis of the morphological characteristics of wire ropes,a multi-channel axial attention mechanism was presented to more efficiently extract and focus on more critical feature of the axial re-gion feature information of wire ropes.Secondly,multi-scale fusion was performed through the cen-tralized feature pyramid module to extract feature information on different scales.The combination of the lightweight MobileNetV3 feature extraction network and the anchor-free frame detection head balances detection speed and accuracy.Experiments carried out on a visible image dataset of wire ropes show that the method improves the accuracy by 1.99%compared to the best performing algorithm among the commonly used target detection algorithms.It is capable of better detecting the damage on the surface of wire ropes.

wire rope damagetarget detectioncentralized feature pyramidaxial attention

王迪、徐兴华、邱少华、王天雨、刘子怡

展开 >

海军工程大学电磁能技术全国重点实验室,武汉 430033

湖北东湖实验室,武汉 430202

华中科技大学电子信息与通信学院,武汉 430074

钢索损伤 目标检测 中心化特征金字塔 轴向注意力

2024

海军工程大学学报
海军工程大学

海军工程大学学报

CSTPCD北大核心
影响因子:0.34
ISSN:1009-3486
年,卷(期):2024.36(6)