摘要
针对目前肉眼检测服装缝线疵点效率低下、成本较高、准确率不高、容易出现漏检和误检等问题,文章利用深度学习的方式对服装缝线疵点进行实时检测,构建了一个服装缝线疵点检测的数据集,包含了常见的服装缝线疵点类型,提出了一种基于注意力机制的YOLOv7算法SK-YOLOv7,在YOLOv7的骨干网络添加3个SK模块,以增强主干网络的特征提取能力,并引入CBAM-YOLOv7和SE-YOLOv7算法进行对比实验。实验结果表明:SK-YOLOv7具有较高的查准率及查全率,平均检测精度也有所提高。SK-YOLOv7相较于CBAM-YOLOv7和SE-YOLOv7在缝线检测上表现更好。文章对数据集采用不同的标记方式进行对比测试,发现对疵点区域进行一次标记的方法会导致特征大量丢失,而对疵点区域内进行分块标记的方法表现出了更好的检测效果。综合实验结果得出,基于注意力机制改进的YOLOv7服装缝线疵点检测方法是可行的,可以较好地推动纺织服装检测行业的发展进步。
Aiming at the problems of low efficiency,high cost,low accuracy,and easy to miss and mis-check in the inspector′s naked eye inspection method,deep learning was used in this paper to detect clothing stitch defects in real time.A dataset for garment stitch fault detection was constructed in this paper,which contains common types of garment stitch faults.In addition,an improved attention mechanism of YOLOv7 algorithm SK-YOLOv7 was proposed,three SK modules were added to the backbone network of YOLOv7 to enhance the feature extraction capability of the backbone network,and CBAM-YOLOv7 and SE-YOLOv7 were introduced for comparison experiments.The experimental results show that SK-YOLOv7 has a higher detection accuracy and improved detection completeness as well as mean average precision.SK-YOLOv7 performs better in stitch detection compared to CBAM-YOLOv7 and SE-YOLOv7.In addition,different marking methods were used to compare the dataset.The method of marking the defective area once resulted in a large number of lost features,while the method of marking the defective area in chunks showed a better detection result.The results of the comprehensive experiments are judged that the proposed method for detecting garment stitch faults is fully feasible and can better promote the development of the textile and garment testing industry.
作者
束方盛
徐增波
鲍禹辰
SHU Fangsheng;XU Zengbo;BAO Yuchen(School of Textiles and Fashion,Shanghai University of Engineering Science,Shanghai 201620,China)
出处
《毛纺科技》
CAS
北大核心
2024年第1期107-115,共9页
Wool Textile Journal
