基于改进Inception ResNet V2网络的可回收垃圾分拣系统

徐丽; 周腊吾; 李高嘉

doi:10.13205/j.hjgc.202404027

基于改进Inception ResNet V2网络的可回收垃圾分拣系统

doi: 10.13205/j.hjgc.202404027

长沙理工大学电气与信息工程学院,长沙 410114

基金项目:

长沙理工大学研究生科研创新项目(CSLGCX23059)

湖南省自然科学基金项目(2018JJ2434)

详细信息

作者简介:
徐丽(2000-),女,硕士研究生,主要研究方向为废物资源化利用、人工智能。2206943539@qq.com

通讯作者:
徐丽(2000-),女,硕士研究生,主要研究方向为废物资源化利用、人工智能。2206943539@qq.com

计量
- 文章访问数: 81
- HTML全文浏览量: 14
- PDF下载量: 3
- 被引次数: 0
出版历程
- 收稿日期: 2023-05-17
- 网络出版日期: 2024-06-01

A RECYCLABLE WASTE SORTING SYSTEM BASED ON AN IMPROVED INCEPTION RESNET V2 NETWORK

School of Electrical and Information Engineering, Changsha University of Technology,Changsha 410114, China

摘要

摘要: 垃圾围城一直是困扰我国城市管理的一大难题。针对可回收垃圾处理过程难度大的问题,提出了1套基于改进的Inception ResNet V2网络结合ROBOT MG400机械臂进行可回收垃圾自动化分拣的系统。首先,对MG400机械臂上的夹具进行改进,使之更适用于垃圾抓取;然后,自主创建了50850张数据集,在此基础上对垃圾图像经过背景降噪、图像分类以及投票算法的处理,并在Inception ResNet V2网络的输出层加入CBAM注意力机制,提高模型识别的准确性;最后对整个系统进行了实验验证。结果表明:该系统可较为准确地分类垃圾并收集至对应垃圾收集容器中,训练时模型的准确率为99.35%,在系统中传送带运行时识别准确率为95.39%,改进的网络在实际应用中的mAP值比原模型高2.56%,并且系统的分拣效率可达到60件/min。该系统可高效率、高准确率、高精度独立地完成可回收垃圾的分拣工作。
- Inception ResNet V2 /
- 图像分类 /
- 可回收垃圾 /
- 深度学习 /
- 注意力机制
Abstract: Garbage siege has always been a big problem in China’s urban management. Aiming at the difficulties of the recyclable waste disposal process, this paper proposed a system based on the improved Inception ResNet V2 network combined with the ROBOT MG400 robotic arm for automatic sorting of recyclable waste. Firstly, we improved the fixture on the MG400 robotic arm to make it more suitable for garbage grabbing. Then, we independently created a dataset of 50850 sheets, based on which the garbage images were processed by background noise reduction, image classification, and voting algorithms, and the CBAM attention mechanism was added to the output layer of the Inception ResNet V2 network to improve the accuracy of model recognition. Finally, the whole system was experimentally verified. The verification results showed that the system could classify garbage more accurately and collect it into the corresponding garbage collection container, and the recognition accuracy was 99.35% and 95.39% when the conveyor belt was stationary and running, the mAP value of the improved network was 2.56% higher than that of the original model, and the sorting efficiency of the system reached 60 pieces per minute. Therefore, this system can independently complete the sorting of recyclable waste with high efficiency, high accuracy, and high precision.
- Inception ResNet V2 /
- image classification /
- recyclables /
- sorting systems /
- attention mechanisms

HTML全文

参考文献(35)

[1]	贺嘉妮,刘意立,李竺霖,等. 生活垃圾分类运输能耗分析[J]. 环境工程,2021,39(10):136-142.
[2]	任中山,陈瑛,王永明,等. 生活垃圾分类对垃圾焚烧发电产业发展影响的分析[J]. 环境工程,2021,39(6):150-153 ,206.
[3]	何汶峰,郑宇,刘蓓蓓,等. 垃圾分类政策对垃圾焚烧大气污染排放的影响[J].中国环境科学,2022,42(5):2433-2441.
[4]	金宜英,邴君妍,罗恩华,等. 基于分类趋势下的我国生活垃圾处理技术展望[J]. 环境工程,2019,37(9):149-153 ,130.
[5]	袁建野,南新元,蔡鑫,等. 基于轻量级残差网路的垃圾图片分类方法[J]. 环境工程,2021,39(2):110-115.
[6]	张月文,李松恒,张炜,等. 基于机器视觉的可回收垃圾智能分拣系统设计[J]. 实验室研究与探索,2022,41(7):98-103 ,107.
[7]	ABEYWICKRAMA T, CHEEMA M A, TANIAR D. K-nearest neighbors on road networks: a journey in experimentation and in-memory implementation[J]. Proceedings of the Vldb Endowment, 2016, 9(6): 492-503.
[8]	康庄,杨杰,郭濠奇. 基于机器视觉的垃圾自动分类系统设计[J]. 浙江大学学报(工学版),2020,54(7):1272-1280,1307.
[9]	SUN Q R, LIU Y Y, CHEN Z Z, et al. Meta-transfer learning through hard tasks[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2022,44(3): 1443-1456.
[10]	XIE W Z, LI S P, XU W, et al. Study on the CNN model optimization for household garbage classification based on machine learning[J]. Journal of Ambient Intelligence and Smart Environments,2022,14(6):439-454.
[11]	梅志敏,陈艳,胡杭,等. 机器人与机器视觉的垃圾分拣系统设计[J]. 机械设计与制造,2022(4):275-278.
[12]	ZHAO Y, HUANG H C, LI Z X, et al. Intelligent garbage classification system based on improve MobileNetV3-Large[J]. Connection Science,2022,34(1).
[13]	程镕杰,杨耘,李龙威,等. 基于深度可分离卷积的轻量化残差网络高光谱影像分类[J]. 光学学报,2023,43(12):303-312.
[14]	张栋,姜媛媛. 融合注意力机制与逆残差结构的轻量级钻机目标检测方法[J]. 电子测量与仪器学报,2022,36(11):201-210.
[15]	卢鹏,曹阳,邹国良,等. 改进Shufflenetv2_YOLOv5的轻量级SAR图像舰船目标实时检测[J].海洋测绘,2023,43(1):58-62 ,82.
[16]	LIN K S, ZHOU T, GAO X F, et al. Deep convolutional neural networks for construction and demolition waste classification: VGGNet structures, cyclical learning rate, and knowledge transfer[J]. Journal of Environmental Management,2022,318:115501.
[17]	HAYAT M, BENNAMOUN M, AN S J. Deep reconstruction models for image set classification[J]. IEEE transactions on pattern analysis and machine intelligence,2015,37(4).
[18]	REN S Q, HE K M, 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.
[19]	SZEGEDY C, IOFFE S,VANHOUCKE V.Inception-v4,lnception-ResNet and the impact of residual connections on learning[J].arXiv:1602.07261,2016.
[20]	刘星辰,周奇才,赵炯,等. 一维卷积神经网络实时抗噪故障诊断算法[J]. 哈尔滨工业大学学报,2019,51(7):89-95.
[21]	SZEGEDY C,LOFFE S,VANHOUCKE V, et al.Inception-v4,inception-resnet and the impact of residual connections onlearning[C]//Proceedings of Association for the Advance of Artificial Intelligence.Menlo Park:AAAI,2017:4-12.
[22]	刘星辰,周奇才,赵炯,等. 一维卷积神经网络实时抗噪故障诊断算法[J]. 哈尔滨工业大学学报,2019,51(7):89-95.
[23]	BUTERA L,FERRANTE A,JERMINI M, et al. Precise agriculture: effective deep learning strategies to detect pest insects[J].IEEE/CAA Journal of Automatica Sinica,2022,9(2):246-258.
[24]	HE K M, ZHANG X Y, REN S Q, et al.Deep residual learning for image recognition[C]//29th lEEE Conference on Computer Vision and Pattern Recognition.Las Vegas:lEEE Computer Society,2016:770-778.
[25]	马燕,余海军,钟发生,等. 基于残差编解码网络的CT图像金属伪影校正[J]. 仪器仪表学报,2020,41(8):160-169.
[26]	范敏,孟鑫余,夏嘉璐,等. 云边协同下基于深度迁移网络的配电台区异常工况诊断方法[J]. 电机与控制学报,2023,27(1):128-138.
[27]	肖鹏程,徐文广,张妍,等. 基于SE注意力机制的废钢分类评级方法[J]. 工程科学学报,2023,45(8):1342-1352.
[28]	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.
[29]	樊继慧,滕少华,金弘林. 基于改进Sigmoid卷积神经网络的手写体数字识别[J]. 计算机科学,2022,49(12):244-249.
[30]	肖蔓君,陈思颖,倪国强,等. 基于Sigmoid函数局部视觉适应模型的真实影像再现[J]. 光学学报,2009,29(11):3050-3056.
[31]	王绍文,宋鹏,谭军,等. 基于QHAdam梯度优化算法的最小二乘逆时偏移[J]. 地球物理学报,2022,65(7):2673-2680.
[32]	陈纪宏,卞荣星,张听雪,等. 垃圾分类对碳减排的影响分析:以青岛市为例[J].环境科学,2023,44(5):2995-3002.
[33]	张天,孙连英,杨琰,等. 基于改进残差网络的陷落柱识别方法[J]. 煤田地质与勘探,2023,5(5):171-179.
[34]	张洪,盛永健,黄子龙,等. 基于W-DenseNet的减压阀不平衡样本故障诊断模型[J]. 控制与决策,2022,37(6):1513-1520.
[35]	白中浩,李智强,蒋彬辉,等. 基于改进YOLOv2模型的驾驶辅助系统实时行人检测[J]. 汽车工程,2019,41(12):1416-1423.