基于DCA-CenterNet的棉布瑕疵检测研究  

Research on cotton fabric defect detection based on DCA-CenterNet

作　　者：苏莲花 李波[1] 杨正达 姚为[1] SU Lianhua;LI Bo;YANG Zhengda;YAO Wei(College of Computer Science,South-Central Minzu University,Wuhan 430074,China)

机构地区：[1]中南民族大学计算机科学学院,武汉430074

出　　处：《中南民族大学学报（自然科学版）》2024年第5期683-691,共9页Journal of South-Central University for Nationalities：Natural Science Edition

基　　金：国家自然科学基金资助项目(61976226)。

摘　　要：针对棉布生产过程中常存在各类瑕疵,且瑕疵的尺度差异较大、形态各异、对比度低、部分为小目标瑕疵等特点,提出了一种基于CenterNet网络的改进目标检测模型DCA-CenterNet,并首次应用于棉布瑕疵检测.在骨干网络Hourglass的残差模块引入CA注意力机制,可以捕捉到不同位置之间的空间关系和上下文信息,提高了网络对于棉布瑕疵的特征表达能力;设计了基于定位质量的关键点筛选模块,可以有效地捕获关键位置信息,提高算法模型检测精度;采用多组改进的带有基于定位质量的关键点筛选模块的检测器,以更好地适应棉布瑕疵的种类多样性和尺度差异,有效解决棉布瑕疵极端长宽比问题.实验结果表明:提出的模型相较于改进前的模型,mAP提高了4.14%,比YOLOv5、FasterRCNN算法分别高出了4.20%和9.11%,验证了所提模型的有效性.Addressing various types of defects commonly found in the cotton fabric production process,and considering the significant differences in scale,diverse shapes,low contrast,and small object defects,an improved object detection model,DCA-CenterNet,based on the CenterNet network is proposed.This model is applied to detect defects in cotton fabric for the first time.By introducing the Coordinate Attention into the residual module of the backbone network Hourglass,the model can capture spatial relationships and contextual information of different positions,enhancing the feature expression capability for cotton fabric defects.Key point filtering module is designed to effectively capture crucial position information,thereby improving the algorithm model's detection accuracy.Multiple groups of improved detectors with key point filtering modules based on localization quality are employed to better adapt to the diversity and scale differences of cotton fabric defects,effectively addressing the issue of extreme aspect ratios in cotton fabric defects.Experimental results demonstrate that the proposed model achieves a 4.14%improvement in mAP compared to the original model.Furthermore,it outperforms YOLOv5 and Faster RCNN algorithms by 4.20% and 9.11%,respectively.These validate the effectiveness of the proposed model.

关 键 词：DCA-CenterNet算法 棉布瑕疵 瑕疵检测 Hourglass网络 小目标检测 注意力机制 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]