基于条件GAN与自适应融合的胶带表面缺陷数据构建方法  

Construction method of adhesive tape surface defect data based on conditional GAN and adaptive fusion

作　　者：崔志芳 洪文颖 武宏旺 冀杰 CUI Zhi-fang;HONG Wen-ying;WU Hong-wang;JI Jie(Shanxi Xinyuan Coal Co.,LTD.,Shouyang,Shanxi 045000,China)

机构地区：[1]山西新元煤炭有限责任公司,山西寿阳045000

出　　处：《煤炭加工与综合利用》2024年第7期41-46,共6页Coal Processing & Comprehensive Utilization

摘　　要：针对煤矿胶带输送机在复杂工作环境下容易出现表面缺陷,影响生产和安全的实际问题,采用基于条件对抗生成网络的图像合成算法,结合传统图片处理技术,提高缺陷检测的准确性和稳定性。研究结果表明:自适应融合算法能够生成真实且稳定的合成图片,且生成的样本无需手动标注目标框,有助于快速建立监督学习需要的训练样本集,解决了煤矿环境对算法的挑战。利用数据合成算法,目标检测率提升5%以上。In view of the practical problem that surface defects are prone to occur in coal mine belt conveyor in complex working environment,which affects production and safety,the image synthesis algorithm based on conditional adversity-generating network and traditional image processing technology are adopted to improve the accuracy and stability of defect detection.The research results show that the adaptive fusion algorithm can generate real and stable composite images,and the generated samples do not need to manually mark the target box,which helps to quickly establish the training sample set required for supervised learning,and solves the challenge of the algorithm in the coal mine environment.Using the data synthesis algorithm,the target detection rate is increased by more than 5%.

关 键 词：胶带输送机 GAN 自适应融合 表面缺陷 目标检测率 

分 类 号：TD528.1[矿业工程—矿山机电]