基于改进YOLOv5s的沙滩垃圾识别模型  

Beach Litter Recognition Model Based on Improved YOLOv5s

作　　者：朱颖 夏靖邦 林康 卢颖裕 岑淑桢 郑壹 朱孝斌 Zhu Ying;Xia Jingbang;Lin Kang;Lu Yingyu;Cen Shuzhen;Zheng Yi;Zhu Xiaobin(Guangzhou Huali College 511300,China)

机构地区：[1]广州华立学院,广州511300

出　　处：《机电工程技术》2025年第3期97-101,126,共6页Mechanical & Electrical Engineering Technology

基　　金：2023年广东省科技创新战略专项资金(pdjh2023b0747);2022年广东省大学生创新训练项目(S202213656018)。

摘　　要：沙滩垃圾存在数量增多、人工清洁效率低和成本高等问题,当前许多基于视觉的垃圾检测与分类模型普遍存在小目标检测和特征提取方面的不足。为解决上述分类问题,结合深层卷积神经网络和目标检测算法,尝试将基于深度学习的图像分类算法用于沙滩垃圾识别。以YOLOv5s为基础模型,提出如下改进方案:针对检测精度问题,重新设计网络添加P2检测头,从而加强小目标的检测的能力,并且为了进一步提升网络提取特征的能力,引入CBAM注意力机制,嵌入在CNN卷积层之间。实验数据表明,所提识别模型在不同光照、角度、遮挡和距离等情况下,准确率为92%,相较于未改进的YOLOv5s准确率提高了4%,改进后的模型mAP值由0.849提高到0.898。Garbage left on beach is increasing,but the efficiency of manual cleaning is low and the cost is high.At present,many of the visionbased garbage detection and sorting models generally have shortcomings in small target detection and feature extraction.In order to solve the above sorting problems,combined with deep convolutional neural network and object detection algorithm,the image classification algorithm based on deep learning is applied to beach garbage recognition.Based on the YOLOv5s model,the following improvement scheme is proposed:In order to solve the detection accuracy problem,the network is redesigned and P2 detection header is added to enhance the detection capability of small targets.In addition,in order to further improve the feature extraction capability of the network,CBAM attention mechanism is introduced and embedded between CNN convolutional layers.Experimental data show that under different illumination,angle,occlusion and distance,the accuracy of the proposed recognition model is 92%,which is 4%higher than that of the unimproved YOLOv5s,and the mAP value of the improved model is increased from 0.849 to 0.898.

关 键 词：YOLOv5s CBAM注意力机制 垃圾识别 沙滩垃圾 分类回收 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]