摘要
蚕茧是纺织品重要的原材料,带有缺陷的蚕茧会对纺织品质量造成影响,目前人工检测的效率低,成本高,针对这一问题提出通过深度学习技术提高蚕茧缺陷检测的效率。本文首先通过左下角填充算法构造多蚕茧同时识别数据集,然后利用EfficientDet网络模型对该数据集进行识别,最后通过实验比较EfficientDet-D0、YOLOv5s、SSD神经网络在构建的数据集上的性能,实验结果表明EfficientDet-D0网络模型的mAP为82.4%,优于另外2种算法,具有较好的准确率。
Cocoon is an important raw material for textiles.Defective cocoons affect the quality of textiles.At present,the efficiency of manual detection of defective cocoons is low and the cost is high.In order to solve this problem,our paperproposes to improve the efficiency of defective cocoon detection through deep learning technology.In this paper,the data set of simultaneous recognition of multiple cocoons is constructed by filling the algorithm in the lower left corner,and then the EfficientDet network model is used to recognize the data set.Finally,the performance of efficientdet-d0,Yolo v5s and SSD neural networks on the constructed data set is compared through experiments.The experimental results show that the mAPvalue of EfficientDet-D0 network model is 82.4%,which is better than the other two algorithms due to its high accuracy.
作者
汪小东
叶飞
王启真
李子印
杨娟亚
WANG Xiao-dong;YE Fei;WANG Qi-zhen;LI Zi-yin;YANG Juan-ya(Huzhou Institute of Quality and Technical Supervision and Inspection,Huzhou 313000 Zhejiang,China;China Jiliang University,Hangzhou 310018,China)
出处
《蚕桑通报》
2022年第4期31-34,共4页
Bulletin of Sericulture
基金
浙江省市场监督管理局科研计划项目(20210146)
浙江省市场监督管理局雏鹰计划培育项目(CY2022352)
浙江省基础公益研究计划项目(LGN20F050001)
国家市场监督管理总局科技计划项目(2022MK048)。
