基于深度学习的提花面料检测算法  

A Jacquard Fabric Detection Algorithm Based on Deep Learning

作　　者：李文卓 何梦珂 毋涛[1] LI Wen-zhuo;HE Meng-ke;WU Tao(School of Computer Science,Xi'an Polytechnic University,Xi'an 710048,China;China Textile Industry Federation,Beijing 100020,China)

机构地区：[1]西安工程大学计算机科学学院,陕西西安710048 [2]中国纺织工业联合会,北京100020

出　　处：《计算机技术与发展》2024年第11期193-199,共7页Computer Technology and Development

基　　金：国家自然科学基金面上项目(62176204)。

摘　　要：针对提花面料在人工检测背景下存在的检测耗时和不准确等问题,提出了一种改进的YOLOv5s深度学习算法用于提花面料检测。首先,对YOLOv5s主干网络进行改进,引入EfficientViT网络作为主干特征提取网络,以提高特征提取的效果;其次,为增强网络对于关键区域的特征关注,引入一种动态且具有查询感知的双层路由注意力机制BRA,以减少背景对检测目标的影响;最后,对损失函数进行改进,使用MpdIoU与InnerIoU相结合的损失函数Mpd-InnerIoU替换CIoU,提高边界框的回归精度。实验结果表明,改进后的YOLOv5s算法相较于原算法在提花面料检测任务中取得了显著的性能提升。与原模型相比,算法精度提高了6.5百分点,召回率提高了4.8百分点,mAP@0.5提高了5.3百分点,实验结果表明改进后的算法在各项性能指标上都有一定程度的提高,能够有效地进行提花面料检测的相关任务。In response to the time-consuming and inaccurate nature of manual inspection for jacquard fabrics,an enhanced YOLOv5s deep learning algorithm has been proposed for the detection of such fabrics.Firstly,the main body of the YOLOv5s network was improved by incorporating the EfficientViT network as the backbone for feature extraction,thereby enhancing the effectiveness of feature extraction.Secondly,to enhance the network's focus on key areas,a dynamic and query-aware dual-layer routing attention mechanism,known as BRA,was introduced to reduce the influence of the background on the detection targets.Lastly,the loss function was refined by substituting the CIoU with a combined loss function of MpdIoU and InnerIoU,named Mpd-InnerIoU,to improve the precision of bounding box regression.Experimental results demonstrate that the improved YOLOv5s algorithm has achieved a significant performance improvement over the original algorithm in the task of jacquard fabric detection.Compared to the original model,the algorithm's accuracy has increased by 6.5 percentage points,the recall rate has improved by 4.8 percentage points,and the mAP@0.5 has increased by 5.3 percentage points.The experimental results indicate that the improved algorithm has achieved a certain degree of enhancement in various performance metrics and is capable of effectively performing related tasks of jacquard fabric detection.

关 键 词：提花面料 深度学习 EfficientViT 双层路由注意力 损失函数 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]