基于YOLOx-pro的盖板玻璃复杂缺陷检测方法  

Complex Defect Detection Method for Cover Glass Based on YOLOx-Pro

作　　者：陈湘尹 尹玲 张斐 吴鹏[1] 叶正伟 谷叶阳 CHEN Xiangyin;YIN Ling;ZHANG Fei;WU Peng;YE Zhengwei;GU Yeyang(School of Mechanical Engineering,Dongguan University of Technology,Dongguan 523808,China;Hunan Provincial Key Laboratory of Health Maintenance for Mechanical Equipment,Hunan University of Science and Technology,Xiangtan 411201,China)

机构地区：[1]东莞理工学院机械工程学院,东莞523808 [2]湖南科技大学机械设备健康维护湖南省重点实验室,湘潭411201

出　　处：《组合机床与自动化加工技术》2024年第9期91-97,共7页Modular Machine Tool & Automatic Manufacturing Technique

基　　金：广东省城市生命线工程智慧防灾与应急技术重点实验室项目(2022B1212010016);东莞市科技特派员项目(20211800500252);广东省3C产业智能制造装备创新科研团队项目(2017BT01G167)。

摘　　要：为解决手机盖板玻璃表面复杂缺陷检测精度低、速度慢、检测技术难以部署于应用端的问题,提出一种基于YOLOx-pro模型的快速检测方法。根据盖板玻璃的光学性质,设计打光方案并搭建图像采集系统,结合图像增强技术丰富缺陷样本。以YOLOx-tiny为基础轻量化模型,在主干输出部分添加CA注意力机制,加强对缺陷区域的关注。引入空间池化金字塔SPPF并将激活函数更换为ReLU,构成Sim-SPPF模块,获取更丰富的多尺度信息表达,结合特定的训练策略,进一步提高检测效率。实验结果表明,YOLOx-pro模型的mAP达到85.73%,FPS达到39.17 f/s,而Params仅为10.58 M,性能优于其他主流算法。将模型部署于应用端软件进行实际测试,结果显示YOLOx-pro具备良好的响应速度和准确率,可实现实际工况下盖板玻璃表面缺陷的高效检测。In order to solve the problems of low accuracy and slow speed of detecting complex defects on the surface of mobile phone cover glass and the difficulty of deploying the detection technology on the application side,a fast detection method based on YOLOx-pro model is proposed.According to the optical properties of cover glass,design the lighting scheme and build the image acquisition system,combined with image enhancement technology to enrich the defect samples.The YOLOx-tiny-based lightening model is used,and the CA attention mechanism is added to the main output part to strengthen the attention to the defective region.Introducing the spatial pooling pyramid SPPF and replacing the activation function with ReLU,which constitutes the Sim-SPPF module,acquires richer multi-scale information expression,and combines with specific training strategies to further improve the detection efficiency.The experimental results show that the YOLOx-pro model achieves a mAP of 85.73%and an FPS of 39.17 f/s,while the Params is only 10.58 M,which outperforms other mainstream algorithms.The model is deployed in the application software for actual testing,and the results show that YOLOx-pro has good response speed and accuracy,and can achieve efficient detection of cover glass surface defects under real working conditions.

关 键 词：缺陷检测 YOLOx 注意力机制 空间金字塔池化 模型部署 

分 类 号：TH164[机械工程—机械制造及自动化] TG659[金属学及工艺—金属切削加工及机床]