MNTH-YOLOv8:一种用于食品包装中蚊虫高效检测的深度学习方法  

MNTH-YOLOv8:A Deep Learning Approach for Efficient Mosquito Detection in Food Packaging

作　　者：王晓红[1] 张微 WANG Xiaohong;ZHANG Wei(School of Communication and Art Design,University of Shanghai for Science and Technology,Shanghai 200093,China)

机构地区：[1]上海理工大学出版印刷与艺术设计学院,上海200093

出　　处：《包装学报》2024年第3期91-98,共8页Packaging Journal

基　　金：国家新闻出版署智能与绿色柔版印刷重点实验室招标课题资助(ZBKT202301)。

摘　　要：食品安全一直是社会关注的焦点,而在食品包装印刷生产过程中,蚊虫的夹杂会对食品安全构成威胁。针对食品包装质检过程中蚊虫检测仍是人工筛查的现状,以及蚊虫目标尺寸小、所处背景复杂的特点,提出了一种基于深度学习的全自动MNTH-YOLOv8检测方法。该方法是在YOLOv8强大的目标检测功能基础上,结合通道特征部分卷积模块、SimAM注意力机制和改进的特征融合模块,并以CIoU与归一化Wasserstein距离作为定位回归损失函数的优化模型。对真实数据集的检测结果表明,MNTH-YOLOv8表现出显著优势,不仅有效提高了小目标蚊虫的检测精度,还在保持检测速度的前提下减少了参数量。MNTH-YOLOv8在食品包装中蚊虫的实时检测应用上拥有广阔前景。Food safety has always been a focus of social concern.However,the presence of mosquitoes and other insects during the food packaging and printing process poses a serious challenge to food safety.Aiming at the current situation of manual screening for mosquitoes and other insects during food packaging quality inspection,and account of the small size of insect targets and the complexity of their backgrounds,a fully automatic MNTH-YOLOv8 detection method based on deep learning was proposed.this method was based on the powerful object detection capability of YOLOv8,combined with channel-wise partial convolution modules and SimAM attention mechanism,with CIoU and normalized Wasserstein distance as the localization regression loss function.Experimental results demonstrated significant advantages of the proposed method in real datasets.It not only effectively improved the detection accuracy of small insect targets but also significantly reduced the parameter count while maintaining detection speed,indicating its great prospect in the application of real-time detection of mosquitoes and other insects in food packaging.

关 键 词：食品包装安全 蚊虫检测 YOLOv8 小目标检测 SimAM注意力机制 特征融合 归一化Wasserstein距离 

分 类 号：TB484.1[一般工业技术—包装工程] TP317.4[自动化与计算机技术—计算机软件与理论]