基于多通道轴向注意力的钢索表面损伤视觉检测方法

Visual detection of surface damage on wire ropes based on multi-channel axial attention

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。

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.