基于改进YOLO-V5深度学习模型的靶丸快速筛选方法  

Rapid screening method of ICF capsule based on improved YOLO-V5 deep learning model

作　　者：刘一郡 赵维谦[1] 刘子豪 罗杰 李兆宇 王允[1] LIU Yijun;ZHAO Weiqian;LIU Zihao;LUO Jie;LI Zhaoyu;WANG Yun(MIIT Key Laboratory of Complex-field Intelligent Exploration,School of Optics and Photonics,Beijing Institute of Technology,Beijing 100081,China)

机构地区：[1]北京理工大学光电学院“复杂环境智能感测技术”工信部重点实验室,北京100081

出　　处：《光学技术》2023年第5期591-595,共5页Optical Technique

基　　金：高性能激光差动共焦拉曼光谱显微成像关键技术及系统研究项目(U22A6006);机械测试理论、方法与技术项目(51825501)。

摘　　要：针对激光惯性约束核聚变实验中海量靶丸筛选效率低的问题,提出一种基于改进YOLO-v5深度学习模型的靶丸快速筛选方法。方法通过控制靶丸在不同的景深处成像,并将图像拼接在一起以获得其清晰图像;同时引入通道注意力机制来增强模型的特征提取能力,建立了SE-YOLOV5s深度学习靶丸表面缺陷识别模型,并对靶丸缺陷按照缺陷种类进行了分类和评估从而实现对海量靶丸的筛选。靶丸表面缺陷检测的准确率为94.4%,每秒可检测到约50张靶丸图像(分辨率3072×4096),为激光惯性约束核聚变试验提供一种快速、准确筛选海量靶丸的方法。In order to solve the problem of low efficiency of massive capsules'screening in laser inertial confinement fusion experiments,a rapid capsules screening method based on improved YOLO-V5 deep learning model was proposed.In this method,the capsules were imaged in different scene depths,and the images were spliced together to obtain the clear images;At the same time,the channel attention mechanism was introduced to enhance the feature extraction ability of the model,and the SE-YOLOV5s deep learning capsule surface defects recognition model is established,and the capsule defects are classified and evaluated according to the defect types to achieve the screening of massive capsules.The accuracy of capsule surface defect detection is 94.4%,with fifty capsule images(resolution 3072×4096)detected per second,providing a fast and accurate method for screening massive targets for laser inertial confinement fusion test.

关 键 词：应用光学 聚变靶丸 目标识别 深度学习 YOLO算法 

分 类 号：TP394.1[自动化与计算机技术—计算机应用技术] TH691.9[自动化与计算机技术—计算机科学与技术]