轻量化网络LW-GCNet在垃圾分类中的应用

夏景明; 徐子峰; 谈玲

doi:10.13205/j.hjgc.202302023

轻量化网络LW-GCNet在垃圾分类中的应用

doi: 10.13205/j.hjgc.202302023

1. 南京信息工程大学人工智能学院, 南京 210044;
2. 南京信息工程大学数字取证教育部工程研究中心, 南京 210044

基金项目:

国家自然科学基金项目(61871032)

国家重点研发计划(2021YFB2700910)

江苏省高等学校基金项目(20KJB510036)

详细信息

作者简介:
夏景明(1980-),男,副教授,主要研究方向为图像分类和目标检测。xiajingming@nuist.edu.cn

通讯作者:
谈玲(1979-),女,教授,主要研究方向为大数据和深度学习。001071@nuist.edu.cn

计量
- 文章访问数: 205
- HTML全文浏览量: 32
- PDF下载量: 5
- 被引次数: 0
出版历程
- 收稿日期: 2022-05-02
- 网络出版日期: 2023-05-25
- 刊出日期: 2023-02-01

APPLICATION RESEARCH OF LIGHTWEIGHT NETWORK LW-GCNet IN GARBAGE CLASSIFICATION

1. School of Artificial Intelligence, Nanjing University of Information Science & Technology, Nanjing 210044, China;
2. Engineering Research Center of Digital Forensics Ministry of Education, Nanjing University of Information Science & Technology, Nanjing 210044, China

摘要

摘要: 垃圾分类是构建绿色城市的重要途径。传统的垃圾分类是由人工进行,分类不彻底,工作强度大,不利于环境保护与资源再利用。为提高垃圾分类的准确性,提出了一种基于VGG16网络的轻量化网络模型LW-GCNet (light weight garbage classify network)。该网络模型通过引入深度可分离卷积和SE(squeeze-and-excitation)模块来进行特征提取,并将垃圾图像的浅层和深层特征有机融合,在减少计算量的同时,增强了待分类垃圾图像通道之间的依赖关系,为分类提供多层次的语义信息。此外,LW-GCNet模型采用自适应最大池化和全局平均池化取代VGG16网络中的全连接层,有效降低了参数量。利用由4类垃圾图像构成的数据集GRAB125对LW-GCNet性能进行验证。实验结果表明:该方法在保证识别速度的前提下,识别平均准确率达到77.17%,参数量为3.15M,易于在户外的嵌入式系统中进行部署。
- 垃圾分类 /
- 轻量化网络 /
- SE模块 /
- 特征融合
Abstract: Garbage classification is an important way to build a green city. The traditional garbage classification is commonly carried out manually, the classification is not thorough, and the labor intensity classification is high, which is not conducive to environmental protection and resource reuse. In order to improve the accuracy of garbage classification, this paper proposed a lightweight network model LW-GCNet (light weight garbage classify network) based on VGG16. The network model performed feature extraction by introducing depthwise separable convolution and SE (squeeze-and-excitation) modules, and organically fused the shallow and deep features of junk images. These modules enhanced the dependencies between channels of garbage images to be classified while reducing the computational complexity of the model and providing multi-level semantic information for accurate classification. In addition, the LW-GCNet model adopted adaptive max pooling and global average pooling to replace the fully connected layer in the VGG16 network, which effectively reduced the number of parameters. The performance of LW-GCNet was validated using the dataset GRAB125 consisting of four types of garbage images. The experimental results showed that, on the premise of ensuring the recognition speed, the average recognition accuracy rate of this method reached 77.17%, and the parameter quantity was 3.15 M, making it easy to be deployed in outdoor embedded systems.
- garbage classification /
- lightweight network /
- squeeze-and-excitation module /
- feature fusion

HTML全文

参考文献(28)

[1]	孙晓杰, 王春莲, 李倩, 等. 中国生活垃圾分类政策制度的发展演变历程[J]. 环境工程, 2020, 38(8):65-70.
[2]	任中山, 陈瑛, 王永明. 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.
[3]	王肇嘉, 秦玉, 顾军, 等. 生活垃圾焚烧飞灰二噁英控制技术研究进展[J]. 环境工程, 2021,39(10):116-123.
[4]	KRIZHEVSKY A, SUTSKEVER I, HINTON G E. Imagenet classification with deep convolutional neural networks[J]. Advances in neural information processing systems, 2012, 25:1097-1105.
[5]	SIMONYAN K, ZISSERMAN A. Very deep convolutional networks for large-scale image recognition[J]. arXiv:1409.1556, 2014.
[6]	SZEGEDY C, IOFFE S, VANHOUCKE V, et al. Inception-v4, inception-resnet and the impact of residual connections on learning[C]//The thirty-first AAAI conference on artificial intelligence. 2017.
[7]	TAN M X, LE Q V. Efficientnet:rethinking model scaling for convolutional neural networks[C]//the International Conference on Machine Learning. PMLR, 2019:6105-6114.
[8]	YANG M, THUNG G. Classification of trash for recyclability status[R]. CS229 Project Report, 2016, 2016:3.
[9]	ARAL R A, KESKIN R, KAYA M, et al. Classification of trashnet dataset based on deep learning models[C]//2018 IEEE International Conference on Big Data (Big Data). IEEE, 2018:2058-2062.
[10]	RABANO S L, CABATUAN M K, SYBINGCO E, et al. Common garbage classification using mobilenet[C]//2018 IEEE 10th International Conference on Humanoid, Nanotechnology, Information Technology, Communication and Control, Environment and Management (HNICEM). IEEE, 2018:1-4.
[11]	KENNEDY T. OscarNet:Using transfer learning to classify disposable waste[R]. CS230 Report:Deep Learning. Stanford University, CA, Winter, 2018.
[12]	OZKAYA U, SEYFI L. Fine-tuning models comparisons on garbage classification for recyclability[J]. arXiv:1908.04393, 2019.
[13]	KANG Z, YANG J, LI G, et al. An automatic garbage classification system based on deep learning[J]. IEEE Access, 2020, 8:140019-140029.
[14]	SHI C P, XIA R Y, WANG L G. A novel multi-branch channel expansion network for garbage image classification[J]. IEEE Access, 2020, 8:154436-154452.
[15]	ZENG M, LU X Z, XU W K, et al. PublicGarbageNet:a deep learning framework for public garbage classification[C]//2020 39th Chinese Control Conference (CCC). IEEE, 2020:7200-7205.
[16]	LI Y F, LIU W. Deep learning-based garbage image recognition algorithm[J]. Applied Nanoscience, 2021:1-10.
[17]	MITTAL G, YAGNIK K B, GARG M, et al. Spotgarbage:smartphone app to detect garbage using deep learning[C]//Proceedings of the 2016 ACM International Joint Conference on Pervasive and Ubiquitous Computing, 2016:940-945.
[18]	PROENÇA P F, SIMÕES P. TACO:Trash annotations in context for litter detection[J]. arXiv:2003.06975, 2020.
[19]	PANWAR H, GUPTA P K, SIDDIQUI M K, et al. AquaVision:automating the detection of waste in water bodies using deep transfer learning[J]. Case Studies in Chemical and Environmental Engineering, 2020, 2:100026.
[20]	GUO J B, LI Y X, LIN W Y, et al. Network decoupling:from regular to depthwise separable convolutions[J]. arXiv:1808.05517, 2018.
[21]	CHOLLET F. Xception:Deep learning with depthwise separable convolutions[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2017:1251-1258.
[22]	ZHAO Q T, SHENG T, WANG Y T, et al. M2det:A single-shot object detector based on multi-level feature pyramid network[C]//Proceedings of the AAAI Conference on Artificial Intelligence, 2019, 33(1):9259-9266.
[23]	HAASE D, AMTHOR M. Rethinking depthwise separable convolutions:How intra-kernel correlations lead to improved mobilenets[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2020:14600-14609.
[24]	GUO Y H, LI Y D, WANG L Q, et al. Depthwise convolution is all you need for learning multiple visual domains[C]//Proceedings of the AAAI Conference on Artificial Intelligence, 2019, 33(1):8368-8375.
[25]	HUA B S, TRAN M K, YEUNG S K. Pointwise convolutional neural networks[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2018:984-993.
[26]	HU J, SHEN L, SUN G. Squeeze-and-excitation networks[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2018:7132-7141.
[27]	KAR A, RAI N, SIKKA K, et al. Adascan:adaptive scan pooling in deep convolutional neural networks for human action recognition in videos[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2017:3376-3385.
[28]	LIN M, CHEN Q, YAN S. Network in network[J]. arXiv:1312.4400, 2013.