基于轻量级残差网路的垃圾图片分类方法

袁建野; 南新元; 蔡鑫; 李成荣

doi:10.13205/j.hjgc.202102017

基于轻量级残差网路的垃圾图片分类方法

doi: 10.13205/j.hjgc.202102017

1. 新疆大学电气工程学院, 乌鲁木齐 830047;
2. 中国科学院自动化研究所智能制造技术与系统研究中心, 北京 100190

详细信息

作者简介:
袁建野(1993-),男,硕士研究生,主要研究方向为深度学习,可回收垃圾图像处理。aku_cintapadamu@163.com

通讯作者:
南新元(1967-),男,博士,教授,主要研究方向为机器学习,工业环境技术与应用。nxyxd@sina.com

计量
- 文章访问数: 389
- HTML全文浏览量: 104
- PDF下载量: 16
- 被引次数: 0
出版历程
- 收稿日期: 2020-04-06
- 网络出版日期: 2021-07-19

GARBAGE IMAGE CLASSIFICATION BY LIGHTWEIGHT RESIDUAL NETWORK

1. College of Electrical Engineering, Xinjiang University, Urumqi 830047, China;
2. Intelligent Manufacturing Technology and System Research Center, Institute of Automation, Chinese Academy of Sciences, Beijing 100190, China

摘要

摘要: 近年来，我国生活垃圾总量每年以10%的速度增长。但是生活垃圾的分类处理能力及技术相对比较有限和落后。基于机器视觉的分类方法一直是广泛使用的方法，而传统的视觉分类网络目前面临着参数多、计算量大、分类精度不高和分类时间长的问题。因此，提出使用最大平均组合池化（Max-AVE Pooling）代替ResNet-50Bottleneck中的最大池化（Max Pooling）与平均池化（AVE Pooling）；通过使用深度可分离卷积代替ResNet-50Bottleneck中的标准卷积的方法对垃圾图片进行分类。实验结果表明，该轻量级残差网路（MaxAVE-Pooling-MobileNet-18，MAPMobileNet-18）与经典分类网络相比能使参数量显著减少10倍，计算量显著减少14倍，又能略微提高精度，非常适合移动手机端、嵌入式设备的实施处理和应用。
- 垃圾分类 /
- 残差网络 /
- 深度可分离卷积 /
- 轻量级网络
Abstract: In recent years, the total domestic garbage in China has been increasing at a rate of 10% per year. However, the technology for the classification and treatment of domestic garbage is relatively limited and backward. The classification method based on machine vision has been widely used. Traditional visual classification net currently faces the problems of sophisticated parameters, large amount of calculation, low classification accuracy, and long classification time. Therefore, this paper proposed to use Max-AVE Pooling instead of Max Pooling or AVE Pooling in ResNet-50Bottleneck, and use the depth separable convolution instead of the standard convolution method in ResNet-50Bottleneck to classify junk images. The experimental results showed that the lightweight residual network (MaxAVE-Pooling-MobileNet-18, MAPMobileNet-18) proposed in this paper could significantly reduce the parameter amount by 10 times and the calculation amount by 14 times, and slightly improve the accuracy compared with the classical classification network. It is very suitable for the implementation and application of mobile phones and embedded devices.
- garbage classification /
- residual network /
- deep separable convolution /
- lightweight net

HTML全文

参考文献(21)

[1]	张英民,尚晓博,李开明,等.城市生活垃圾处理技术现状与管理对策[J].生态环境学报,2011,20(2):389-396.
[2]	孟菁华,刘辉,史学峰,等.垃圾焚烧设施居民暴露吸入性健康风险评价研究[J].环境工程,2018,36(1):128-133.
[3]	孙思明.上海市垃圾分类政策执行过程研究[J].现代经济信息,2019(34):496.
[4]	"全国垃圾分类"小程序正式上线[J].城乡建设,2019(24):5.
[5]	吴健,陈豪,方武.基于计算机视觉的废物垃圾分析与识别研究[J].信息技术与信息化,2016(10):81-83.
[6]	黄惠玲,韩军,吴飞斌,等.建筑垃圾的颜色特征提取与分类研究[J].光学与光电技术,2018,16(1):53-57.
[7]	吴碧程,邓祥恩,张子憧,等.基于卷积神经网络的智能垃圾分类系统[J].物理实验,2019,39(11):44-49.
[8]	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.
[9]	LECUN Y,BOSER B,DENKER J, et al.Backpropagation applied to handwritten zip code recognition[J].Neural Computation,1989,1(4):541-551.
[10]	KRIZHEVSKY A,SUTSKEVER I,HINTON G.ImageNet Classification with Deep Convolutional Neural Networks[C]//NIPS,Curran Associates Inc,2012.
[11]	SIMONYAN K, ZISSERMAN A. Very deep convolutional networks for large-scale image recognition[J]. arXiv,2014, 1409.1556.
[12]	SZEGEDY, C, LIU W, JIA Y Q, et al. Going deeper with convolutions[J]. The IEEE Conference on Computer Vision and Pattern Recognition (CVPR). 1-9. DOI: 10.1109/CVPR.2015.7298594.
[13]	HE K M, ZHANG X Y, REN S Q, et al.(2016). Deep Residual Learning for Image Recognition[J]. 770-778. DOI: 10.1109/CVPR.2016.90.
[14]	IANDOLA, F N, HAN S MOSKEWIQ M W, et al. SqueezeNet:alexNet-level accuracy with 50x fewer parameters and<0.5MB model size[J].
[15]	CHOLLET F. Xception:deep learning with depthwise separable convolutions[J]. 1800-1807. DOI: 10.1109/CVPR.2017.195.
[16]	ZHANG X Y, ZHOU X Y, LIN M X, et al. ShuffleNet:an extremely efficient convolutional neural network for mobile devices[J]. 6848-6856. DOI: 10.1109/CVPR.2018.00716.
[17]	YU D J, WANG H L, CHEN P Q, et al. Mixed pooling for convolutional neural networks[J]. 364-375. DOI: 10.1007/978-3-319-11740-9_34.
[18]	HOWARD A G, ZHU M L, CHEN B, et al. MobileNets:efficient convolutional neural networks for mobile vision applications[C].2017.
[19]	HUA, B S,TRAN M K, YEUNG S K. Pointwise convolutional neural networks[J].984-993. DOI: 10.1109/CVPR.2018.00109.
[20]	HOWARD, ANDREW, PANG, et al. Searching for MobileNetV3[J]. 1314-1324. DOI: 10.1109/ICCV.2019.00140.
[21]	HUANG G, LIU Z, PLEISSG, et al. Convolutional Networks with Dense Connectivity[J].Research Gate, 2020.