当前位置>主页 > 期刊在线 > 计算机技术 >

计算机技术22年17期

基于 SqueezeNet-Tiny 的可回收垃圾智能 垃圾桶设计及实现
李长江¹,余海涛 ²,李官星³,李玥洋²,谭雯文²
(1. 重庆科技学院 机械与动力工程学院,重庆 401331;2. 凤鸣山中学,重庆 400030;3. 重庆工程职业技术学院,重庆 402260)

摘  要:文章提出一种对 SqueezeNet 神经网络改进策略,通过删除第三个 Maxpooling 层,同时将第一个卷积层的卷积核大小设置成 3×3,创建轻量化 SqueezeNet-Tiny 模型,并在具有复杂背景的可回收垃圾数据集(Recyclable Waste Dataset,简称 RW Dataset)上验证了改进的有效性。将 SqueezeNet-Tiny 迁移到硬件设施,完成了智能分类垃圾桶的设计和制作,识别精度可达 94.68%,参数量仅为 0.74 M,基本满足工程化应用的需求。


关键词:SqueezeNet;可回收垃圾;智能分类垃圾桶;工业应用



DOI:10.19850/j.cnki.2096-4706.2022.17.019


中图分类号:TP242                                         文献标识码:A                                     文章编号:2096-4706(2022)17-0075-03


Design and Implementation of Intelligent Garbage Can for Recyclable Garbage Based on SqueezeNet-Tiny

LI Changjiang1, YU Haitao2, LI Guanxing3, LI Yueyang2, TAN Wenwen2

(1.School of Mechanical and Power Engineering, Chongqing University of Science and Technology, Chongqing 401331, China; 2.Fengmingshan Middle School, Chongqing 400030, China; 3.Chongqing Vocational Institute of Engineering, Chongqing 402260, China)

Abstract: This paper proposes an improvement strategy for the SqueezeNet neural network. By deleting the third Maxpooling layer and setting the convolution kernel size of the first convolutional layer to 3 × 3, a lightweight SqueezeNet-Tiny model is created. The effectiveness of the improvement is verified on the Recyclable Waste Dataset (RW Dataset) with complex background. The SqueezeNetTiny is migrated to hardware  facilities, and the design and production of intelligent sorting garbage cans are completed. The recognition accuracy can reach 94.68%, and the parameter quantity is only 0.74 M, which basically meets the needs of engineering applications.

Keywords: SqueezeNet; recyclable garbage; smart sorting garbage can; industrial application


参考文献:

[1] RABANO S L,CABATUAN M K,SYBINGCO E,et al. Common Garbage Classification Using MobileNet [C]//International Conference on Humanoid. Baguio City:IEEE,2018:1-4.

[2] HOWARD A G,ZHU M L,CHEN B,et al. MobileNets:Efficient Convolutional NeuralNetworks for Mobile Vision Applications [J/OL].arXiv:1704.04861 [cs.CV].(2017-04-17).https://arxiv.org/ abs/1704.04861.

[3] VICTORIA R,ANGEL S,JOSe F V. Automatic Image-Based Waste Classification [C]//Bioinspired Systems and Biomedical Applications to Machine Learning(IWINAC).Springer:2019:422-431.

[4] IANDOLA F,HAN S,MOSKEWICZ M W,et al. SqueezeNet:AlexNet-level accuracy with 50x fewer parameters and <0.5MB model size [J/OL]. arXiv:1602.07360 [cs.CV].(2016-11-04).https://arxiv.org/abs/1602.07360.

[5] 陈智超,焦海宁,杨杰,等 . 基于改进 MobileNet v2 的垃圾图像分类算法 [J]. 浙江大学学报(工学版),2021,55(8):1490-1499.

[6] 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.

[7] 金佩薇,姚燕,梁晓瑜,等 . 垃圾图像识别研究进展 [J].环境工程,2022,40(1):196-206.

[8] 刘小兵 . 小模数齿轮视觉测量若干关键问题研究 [D]. 北京:北京工业大学,2020.

[9] 吴晓玲,黄金雪,何文海 . 基于深度卷积神经网络的塑料垃圾分类研究 [J]. 塑料科技,2020,48(4):86-89.

[10] 裘祖荣,石照耀,李岩 . 机械制造领域测量技术的发展研究 [J]. 机械工程学报,2010,46(14):1-11.


作者简介:李长江(1970.12—),男,汉族,四川威远人,教授,博士,研究方向:固废处理;计算机视觉。