摘要
为准确检测出布料表面微小损伤如缺维、浆斑、胶斑等问题,提出基于改进YOLO-DETR的布料表面微小损伤检测方法。在改进YOLO-DETR模型中增加残差模块并扩展特征图尺度,以减少漏检可能性。在模型中融入了CBAM注意力机制和Anchor box,采用K-means聚类算法,通过选取K个聚类中心,并基于距离将其他目标划分到最近的聚类中心,形成K个群组。在迭代过程中,通过最小化群内距离和最大化群间距离,优化聚类结果,采用复合损失函数进行检测,增强模型在复杂环境下的识别能力。实验结果表明,该方法检测精度在90%以上,在面对缺维、浆斑、胶斑等复杂性问题时也有较好的检测能力。
To accurately detect minor damage on the surface of fabrics,such as lack of dimensionality,slurry spots and glue spots,an improved YOLO-DETR based method for detecting minor damage on the surface of fabrics is proposed.In the improved YOLO-DETR model,adding residual modules and expanding feature map scales can reduce the possibility of missed detections.CBAM attention mechanism and Anchor box are integrated into the model.K-means clustering algorithm is adopted.By selecting K cluster centers,other objects are classified into the nearest cluster center according to the distance,and then K groups are formed.During the process of iteration,the clustering results are optimized by minimizing intra group distance and maximizing inter group distance.A compound loss function is used for detection to enhance the recognition ability of the model in complex environments.The experimental results show that the detection accuracy of this method is over 90%,and it also has good detection ability in the face of complex situations,such as lack of dimensionality,slurry spots,and glue spots.
作者
胡玉恒
吴谨
HU Yuheng;WU Jin(School of Information Science and Engineering,Wuhan University of Science and Technology,Wuhan 430080,China)
出处
《现代电子技术》
北大核心
2024年第13期160-163,共4页
Modern Electronics Technique
作者简介
胡玉恒(1997-),男,湖北武汉人,硕士,研究方向为深度学习目标检测。;吴谨(1967-),女,湖北武汉人,博士,教授,研究方向为图像处理与模式识别、信号处理与多媒体通信、检测技术与自动化装置等方面。
