GCA-YOLO:一种改进的钢铁表面缺陷检测算法  

GCA-YOLO:an improved steel surface defect detection algorithm

作　　者：唐湘龙 石兰娟 陶利民[1] 刘文浩 TANG Xianglong;SHI Lanjuan;TAO Limin;LIU Wenhao(School of Information Science and Technology,Hangzhou Normal University,Hangzhou 311121,China)

机构地区：[1]杭州师范大学信息科学与技术学院,浙江杭州311121

出　　处：《杭州师范大学学报(自然科学版)》2025年第1期9-18,共10页Journal of Hangzhou Normal University(Natural Science Edition)

基　　金：国家自然科学基金项目(U21A20466).

摘　　要：针对现有钢铁表面缺陷检测算法存在检测精度低、实时性差及参数量大等问题,提出了一种改进的钢铁表面缺陷检测算法——GCA-YOLO.该算法通过嵌入Ghost模块形成C3Ghost结构进行基本特征提取,并用GhostConv替换颈部普通卷积,确保模型检测精度的同时减少模型参数量.在主干网络中融入坐标注意力机制(coordinate attention,CA),增强模型捕获方向和位置信息的敏感度.在下游特征融合网络中引入ACmix模块,集成了自注意力机制和卷积模块,以较低的计算成本提升网络性能.同时,采用SIoU损失函数以提高检测框回归精度.在NEU-DET数据集上的实验结果表明:与原始算法相比,改进算法的平均精度提高10.96百分点,参数量降低了40.68%,仅为4.17 M.对比其他主流目标检测算法,改进算法在精度、速度上均有显著提升,复杂度大幅降低,能够满足钢铁表面缺陷检测的实时性需求.An improved steel surface defect detection algorithm(GCA-YOLO)was proposed to solve the problems of low model detection accuracy,poor real-time performance and large number of parameters in the existing steel surface defect detection algorithms.A C3Ghost structure embedded with Ghost module was designed for basic feature extraction,and GhostConv was used to replace the ordinary neck convolution to ensure the accuracy of the model and reduce the amount of model parameters.Coordinate attention mechanism was integrated into the backbone network to enable the model to capture sensitive information of direction and position.The ACmix module was introduced in the downstream feature fusion network,which integrated the self-attention mechanism and the convolution module,and improved the network performance with lower computational overhead.The SIoU loss function was used to improve the accuracy of detection box regression.The experimental results on NEU-DET dataset showed that the mean average precision of the improved algorithm was 10.96 percentage points higher than that of the original method,and the number of parameters was only 4.17 M,which was reduced by 40.68%.Compared with other mainstream target detection algorithms,the improved algorithm significantly improves the accuracy and speed,and reduces the complexity,which can meet the requirements of real-time steel surface defect detection.

关 键 词：钢铁表面缺陷检测 冗余信息 坐标注意力 特征融合 

分 类 号：TP399[自动化与计算机技术—计算机应用技术]