基于LGD-YOLO高精度轻量化目标检测网络的垃圾检测研究

肖立中; 胡凡

doi:10.13205/j.hjgc.202406020

基于LGD-YOLO高精度轻量化目标检测网络的垃圾检测研究

doi: 10.13205/j.hjgc.202406020

肖立中,
胡凡^,

上海应用技术大学计算机科学与信息工程学院, 上海 201418

基金项目:

上海市自然科学基金项目(20ZR1455600)

详细信息

作者简介:
肖立中(1981-),男,博士,副教授,主要研究方向为软件工程、人工智能。lyexp@163.com

通讯作者:
胡凡(1998-),男,硕士研究生,主要研究方向为软件工程、人工智能。869283597@qq.com

计量
- 文章访问数: 78
- HTML全文浏览量: 8
- PDF下载量: 5
- 被引次数: 0
出版历程
- 收稿日期: 2023-05-23
- 网络出版日期: 2024-07-11

GARBAGE DETECTION BASED ON LGD-YOLO HIGH PRECISION LIGHTWEIGHT OBJECT DETECTION NETWORK

XIAO Lizhong,
HU Fan^,

School of Computer Science and Information Engineering, Shanghai Institute of Technology, Shanghai 201418, China

摘要

摘要: 当前高速发展的社会和迅速增长的城市人口,使得日益严峻的垃圾污染问题越发凸显,垃圾分类处理势在必行。人工处理存在任务重、效率低等问题。部分自动化的分类方法检测精度低,速度慢。为提高复杂场景下垃圾检测准确率,同时轻量化结构,使其便于部署,提出一种融合轻量化卷积模块、注意力机制和多重感受野模块的改进YOLO v5s的垃圾检测模型Lightweight Garbage Detection-YOLO(LGD-YOLO)。首先,在网络结构中引入Ghost卷积和包含GSConv的Slim-Neck模块,使模型变得更加轻量化;其次,嵌入坐标注意力机制,侧重于关注重要信息,以提高检测精度。最后,引入多重感受野模块,提高模型的多尺度检测能力,避免小目标物体的漏检。采用包含不同环境下垃圾图片的Trash_dataset数据集进行测试验证。结果表明:改进后的模型参数量和计算量分别为5.77 M和9.2 GFLOPs,与原模型相比分别减少22.4%和56.4%,单张图片检测速度为26.5 ms,达到垃圾检测的实时性要求。此外,改进的算法具有良好的检测精度,mAP_0.5和mAP_0.5∶0.95分别达到96.20%和77.77%,优于当前流行的目标检测算法。
- 垃圾分类 /
- 目标检测 /
- 轻量化网络 /
- 坐标注意力 /
- 多尺度检测
Abstract: At present, with the rapid development of society and the rapid growth of the urban population, garbage pollution has become increasingly prominent, and garbage classification is imperative. Manual processing has the problems of heavy tasks and low efficiency, and some automated classification methods have low detection accuracy and slow speed. To improve the accuracy of garbage detection in complex scenes, lighten the structure, and make it easier to deploy, an improved YOLO v5s garbage detection model, Lightweight Garbage Detection YOLO (LGD-YOLO) was proposed, which integrated a lightweight convolution module, attention mechanism, and multiple receptive field modules. First, Ghost convolution and Slim Neck module including GSConv were introduced into the network structure to make the model lighter. Secondly, the coordinate attention mechanism was embedded to focus on important information to improve the detection accuracy. Finally, the multi-receptive field module was introduced to improve the multi-scale detection capability of the model and avoid missing detection of small target objects. The data set containing trash garbage images in different environments was tested and verified. The experimental results showed that the parameters and calculation amount of the improved model were 5.77 M and 9.2 GFLOPs, respectively, which were 22.4% and 56.4% less than the original model. The single image detection speed was 26.5 ms, meeting the real-time requirements of garbage detection. In addition, the improved algorithm has good detection accuracy, with mAP_0.5 and mAP_0.5∶0.95 reaching 96.20% and 77.77%, respectively, which was superior to the current popular target detection algorithm.
- garbage classification /
- object detection /
- lightweight network /
- coordinate attention /
- multiscale detection

HTML全文

参考文献(21)

[1]	李金玉,陈晓雷,张爱华,等. 基于深度学习的垃圾分类方法综述[J]. 计算机工程,2022,48(2):1-9.
[2]	LIU W, ANGUELOY D, ERHAN D, et al. SSD: single shot multibox detector[C]//European Conference on Computer Vision, 2016:21-37.
[3]	REDMON J,DIVVALA S,GIRSHICK R, et al. You only look once: unified, real-time object detection[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2016:779-788.
[4]	REDMON J,FARHADI A.YOLOv3: An Incremental Improvement[D].Washington: University of Washington,2018.
[5]	GIRSHICK R, DONAHUE J, DARRELL T, et al. Region-based convolutional networks for accurate object detection and segmentation[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2016, 38(1): 142-158.
[6]	GIRSHICK R. Fast R-CNN[C]//Proceedings of the IEEE International Conference on Computer Vision. 2015:1440-1448.
[7]	REN S, HE K, GIRSHICK R, et al. Faster R-CNN: Towards real-time object detection with region proposal networks[J]. IEEE Transactions on Pattern Analysis & Machine Intelligence, 2017, 39(6):1137-1149.
[8]	王莉,何牧天,徐硕,等.基于YOLOv5s网络的垃圾分类和检测[J].包装工程,2021,42(8):50-56.
[9]	马雯,于炯.基于改进Faster R-CNN的垃圾检测与分类方法[J].计算机工程,2021,47(8):294-300.
[10]	罗安能,万海斌.基于改进YOLOv5s的可回收垃圾的检测算法[J/OL].激光与光电子学进展:1-15[2022-11-15 ].
[11]	HAN K, WANG Y, TIAN Q, et al. GhostNet: More features from cheap operations[C]//2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). IEEE, 2020.
[12]	LI H L, LI J, WEI H B, et al. Slim-neck by GSConv: a better design paradigm of detector architectures for autonomous vehicles[C]//ArXiv abs/2206.02424 (2022): n. pag.
[13]	JIE H,LI S,GANG S, et al. Squeeze-and-excitation networks[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017,99:.
[14]	HOU Q, ZHOU D, FENG J. Coordinate attention for efficient mobile network design[C]//2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR),2021: 13708-13717.
[15]	LIU S T, HUANG D, WANG Y H. Receptive field block net for accurate and fast object detection[C]//European Conference on Computer Vision,2018: 404-419.
[16]	YANG M, THUNG G. Classification of trash for recyclability status[EB/OL]. Stanford University Machine Learning Project Report, CS229, 2016[2021-01-25].
[17]	PROENÇA PEDRO F, PEDRO Simões. TACO: Trash Annotations in Context for Litter Detection[C]//ArXiv abs/2003.06975 (2020): n. pag.
[18]	WOO S, PARK J, LEE J Y, et al. Cbam: convolutional block attention module[C]//Proceedings of the European Conference on Computer Vision (ECCV), 2018: 3-19.
[19]	王一田,唐开强,留沧海,等.基于YOLO v3的地面垃圾检测与清洁度评定方法[J].传感器与微系统,2022(4):129-133.
[20]	袁红春,臧天祺.基于注意力机制Ghost-YOLOv5的水下垃圾目标检测[J].环境工程,2023,41(7):214-221.
[21]	JIANG X, HU H, QIN Y, et al. A real-time rural domestic garbage detection algorithm with an improved YOLOv5s network model[C]//Scientific Reports,2022,12: n. pag.