基于深度学习的图像分类研究综述

doi:10.11959/j.issn.1000-0801.2019268

摘要/Abstract

摘要：

近年来，深度学习在计算机视觉领域中的表现优于传统的机器学习技术，而图像分类问题是其中最突出的研究课题之一。传统的图像分类方法难以处理庞大的图像数据，且无法满足人们对图像分类精度和速度的要求，而基于深度学习的图像分类方法突破了此瓶颈，成为目前图像分类的主流方法。从图像分类的研究意义出发，介绍了其发展现状。其次，具体分析了图像分类中最重要的深度学习方法（即自动编码器、深度信念网络与深度玻尔兹曼机）以及卷积神经网络的结构、优点和局限性。再次，对比分析了方法之间的差异及其在常用数据集上的性能表现。最后，探讨了深度学习方法在图像分类领域的不足及未来可能的研究方向。

关键词: 深度学习, 图像分类, 自编码器, 深度信念网络, 卷积神经网络

Abstract:

In recent years,deep learning performed superior in the field of computer vision to traditional machine learning technology.Indeed,image classification issue drew great attention as a prominent research topic.For traditional image classification method,huge volume of image data was of difficulty to process and the requirements for the operation accuracy and speed of image classification could not be met.However,deep learning-based image classification method broke through the bottleneck and became the mainstream method to finish these classification tasks.The research significance and current development status of image classification was introduced in detail.Also,besides the structure,advantages and limitations of the convolutional neural networks,the most important deep learning methods,such as auto-encoders,deep belief networks and deep Boltzmann machines image classification were concretely analyzed.Furthermore,the differences and performance on common datasets of these methods were compared and analyzed.In the end,the shortcomings of deep learning methods in the field of image classification and the possible future research directions were discussed.

Key words: deep learning, image classification, auto-encoders, deep belief networks, CNN

中图分类号:

TP393

苏赋,吕沁,罗仁泽. 基于深度学习的图像分类研究综述[J]. 电信科学, 2019, 35(11): 58-74.

Fu SU,Qin LV,Renze LUO. Review of image classification based on deep learning[J]. Telecommunications Science, 2019, 35(11): 58-74.

图/表 15

图1

图2

图3

图4

图5

图6

图7

图8

图9

图10

图11

图12

表1

表2

表3

参考文献 86

[1]	OUYANG W , ZENG X , WANG X ,et al. DeepID-Net:object detection with deformable part based convolutional neural networks[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017,39(7): 1320-1334. doi: 10.1109/TPAMI.2016.2587642 pmid: 27392342
[2]	DIBA A , SHARMA V , PAZANDEH A ,et al. Weakly supervised cascaded convolutional networks[C]// IEEE Conference on Computer Vision and Pattern Recognition,July 21-26,2017,Honolulu,HI,USA. New York:ACM Press, 2017: 5131-5139.
[3]	HU G , YANG Y X , YI D ,et al. When face recognition meets with deep learning:an evaluation of convolutional neural networks for face recognition[C]// International Conference on Computer Vision,December 11-18,2015,Santiago,Chile. Piscataway:IEEE Press, 2015: 142-150.
[4]	LAWRENCE S , GILES C L , TSOI A C ,et al. Face recognition:a convolutional neural-network approach[J]. IEEE Transactions on Neural Networks, 1997,8(1): 98-113. doi: 10.1109/72.554195 pmid: 18255614
[5]	CAO Z , SIMON T , WEI S ,et al. Realtime multi-person 2D pose estimation using part affinity fields[C]// IEEE Conference on Computer Vision and Pattern Recognition,July 21-26,2017,Honolulu,HI,USA. EprintArxiv, 2017: 1302-1310.
[6]	TOSHEV A , SZEGEDY C . DeepPose:human pose estimation via deep neural networks[C]// IEEE Conference on Computer Vision and Pattern Recognition,June 23-28,2014,Columbus,OH,USA. New York:ACM Press, 2014: 1653-1660.
[7]	PERREAULT S , HEBERT P . Median filtering in constant time[J]. IEEE Transactions on Image Processing, 2007,16(9): 2389-2394. doi: 10.1109/tip.2007.902329 pmid: 17784612
[8]	SLOT K , KOWALSKI J , NAPIERALSKI A ,et al. Analogue median/average image filter based on cellular neural network paradigm[J]. Electronics Letters, 1999,35(19): 1619-1620. doi: 10.1049/el:19991091
[9]	DIREKOGLU C , NIXON M S . Image-based multiscale shape description using Gaussian filter[C]// 2008 Sixth Indian Conference on Computer Vision,Graphics ＆ Image Processing,December 16-19,2008,Bhubaneswar,India. Piscataway:IEEE Press, 2009: 673-678.
[10]	GRABNER M , GRABNER H , BISCHOF H . Fast approximated SIFT[C]// Asian Conference on Computer Vision,January 13-16,2006,Hyderabad,India. Heidelberg:Springer, 2006: 918-927.
[11]	HE L , ZOU C , ZHAO L ,et al. An enhanced LBP feature based on facial expression recognition[C]// IEEE Engineering in Medicine and Biology 27th Annual Conference,September 1-4,2005,Shanghai,China. Piscataway:IEEE Press, 2005: 3300-3303.
[12]	DENIZ O , BUENO G , SALIDO J ,et al. Face recognition using histograms of oriented gradients[J]. Pattern Recognition Letters, 2011,32(12): 1598-1603. doi: 10.1016/j.patrec.2011.01.004
[13]	LECUN Y , JACKEL L , BOTTOU L ,et al. Comparison of learning algorithms for handwritten digit recognition[C]// International Conference on Artificial Neural Networks,January,1995,Nanterre,France.[S.l.:s.n]. 1995: 53-60.
[14]	BEUCHER A , MOLLER A B , GREVE M H . Artificial neural networks and decision tree classification for predicting soil drainage classes in Denmark[J]. Geoderma, 2017,320: 30-42. doi: 10.1016/j.geoderma.2018.01.018
[15]	EBRAHIMI M A , KHOSHTAGHAZ M H , MINAEI S ,et al. Vision-based pest detection based on SVM classification method[J]. Computers and Electronics In Agriculture, 2017,137: 52-58. doi: 10.3390/s18051489 pmid: 29747429
[16]	周建同, 杨海涛, 刘东 ,等. 视频编码的技术基础及发展方向[J]. 电信科学, 2017,33(8): 16-25.
	ZHOU J T , YANG H T , LIU D ,et al. Trends and technologies of video coding[J]. Telecommunications Science, 2017,33(8): 16-25.
[17]	HINTON G E , SALAKHUTDINOV R R . Reducing the dimensionality of data with neural networks[J]. Science, 2006,313(5786):504. doi: 10.1126/science.1123432 pmid: 16873667
[18]	LECUN Y , BOTTOU L , BENGIO Y ,et al. Gradient-based learning applied to document recognition[J]. Proceedings of the IEEE, 1998,86(11): 2278-2324. doi: 10.1109/5.726791
[19]	XIAO H , RASUL K , VOLLGRAF R . Fashion-MNIST:a novel image dataset for benchmarking machine learning algorithms[J]. Statistics, 2017(2). doi: 10.1080/02331880902986984 pmid: 21243084
[20]	LI H , LIU H , JI X ,et al. CIFAR10-DVS:an event-stream dataset for object classification[J]. Frontiers in Neuroscience, 2017(11):309. doi: 10.3389/fnins.2017.00309 pmid: 28611582
[21]	MCCLURE P , KRIEGESKORTE N . Representational distance learning for deep neural networks[J]. Frontiers in Computational Neuroscience, 2016(10):131. doi: 10.1007/s11548-018-1797-4 pmid: 29850978
[22]	DENG J , DONG W , SOCHER R ,et al. ImageNet:a large-scale hierarchical image database[C]// The 2009 IEEE Conference on Computer Vision and Pattern Recognition,June 20-25,2009,Washington,USA. Piscataway:IEEE Press, 2009: 248-255.
[23]	郭丽丽, 丁世飞 . 深度学习研究进展[J]. 计算机科学, 2015,42(5): 28-33.
	GUO L L , DING S F . Research progress on deep learning[J]. Computer Science, 2015,42(5): 28-33.
[24]	RUMELHART D E , HINTON G E , WILLIAMS R J . Learning representations by back-propagating errors[J]. Nature, 1986,323(6088): 533-536. doi: 10.1038/323533a0
[25]	HINTON G E , OSINDERO S , TEH Y . A fast learning algorithm for deep belief nets[J]. Neural Computation, 2006,18(7): 1527-1554. doi: 10.1162/neco.2006.18.7.1527 pmid: 16764513
[26]	SALAKHUTDINOV R , HINTON G . Deep Boltzmann machines[C]// International Conference on Artificial Intelligence and Statistics,April 16-19,2009,Florida,USA.[S.l.:s.n. ], 2009: 448-455.
[27]	BABRI H A , TONG Y . Deep feedforward networks:application to pattern recognition[C]// International Conference on Neural Networks (ICNN'96),June 3-6,1996,Washington,USA. Piscataway:IEEE Press, 1996: 1422-1426.
[28]	ROSENBLATT F . The perceptron:a probabilistic model for information storage and organization in the brainl[J]. Psychological Review, 1958,65(6): 386-408. doi: 10.1037/h0042519 pmid: 13602029
[29]	赵会敏, 雒江涛, 杨军超 ,等. 集成BP神经网络预测模型的研究与应用[J]. 电信科学, 2016,32(2): 60-67.
	ZHAO H M , LUO J T , YANG J C ,et al. Research and application of prediction model based on ensemble BP neural network[J]. Telecommunications Science, 2016,32(2): 60-67.
[30]	高雪鹏, 丛爽 . BP网络改进算法的性能对比研究[J]. 控制与决策, 2001(2): 167-171.
	GAO X P , CONG S . Comparative study on fast learning algorithms of BP networks[J]. Control and Decision, 2001(2): 167-171.
[31]	OLSHAUSEN B A , FIELD D J . Sparse coding with an overcomplete basis set:A strategy employed by V1?[J]. Vision Research, 1997,37(23): 3311-3325. doi: 10.1016/s0042-6989(97)00169-7 pmid: 9425546
[32]	LIU Y , ZHAO S S , WANG Q Q ,et al. Learning more distinctive representation by enhanced PCA network[J]. Neurocomputing, 2018(275): 924-931.
[33]	LIU T , LI Z R , YU C X ,et al. NIRS feature extraction based on deep auto-encoder neural network[J]. Infrared Physics ＆ Technology, 2017(87): 124-128.
[34]	HASSAIRI S , EJBALI R , ZAIED M . A deep stacked wavelet auto-encoders to supervised feature extraction to pattern classification[J]. Multimedia Tools and applications, 2018,77(5): 5443-5459. doi: 10.1007/s11042-017-4461-z
[35]	LIU Y , WU L Z . Geological disaster recognition on optical remote sensing images using deep learning[J]. Procedia Computer Science, 2016(91): 566-575.
[36]	WANG Y S , YAO H X , ZHAO S C . Auto-encoder based dimensionality reduction[J]. Neuroconmputing, 2016,184(SI): 232-242.
[37]	VINCENT P , LAROCHELLE H , BENGIO Y ,et al. Extracting and composing robust features with denoising autoencoders[C]// the 25th International Conference on Machine Learning,July 5-9,2008,Helsinki,Finland. New York:ACM Press, 2008: 1096-1103.
[38]	VINCENT P , LAROCHELLE H , LAJOIE I ,et al. Stacked denoising autoencoders:learning useful representations in a deep network with a local denoising criterion[J]. Journal of Machine Learning Research, 2010(11): 3371-3408.
[39]	PATHIRAGE C S N , LI J , LI L ,et al. Development and application of a deep learning-based sparse autoencoder framework for structural damage identification[J]. Structural Health Monitoring, 2018,18(1): 103-122. doi: 10.1177/1475921718800363
[40]	LI E Z , DU P J , SAMAT A ,et al. Mid-level feature representation via sparse autoencoder for remotely sensed scene classification[J]. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2017,10(3): 1068-1081. doi: 10.1109/JSTARS.2016.2621011
[41]	RIFAI S , VINCENT P , MULLER X ,et al. Contractive auto-Encoders:explicit invariance during feature extraction[C]// International Conference on Machine Learning,June28-July 2,2011,Washington,USA.[S.l.:s.n]. 2011: 833-840.
[42]	GENG J , FAN J C , WANG H Y ,et al. High-Resolution SAR image classification via deep convolutional autoencoders[J]. IEEE Geoscience and Remote Sensing Letters, 2015,12(11): 2351-2355. doi: 10.1109/LGRS.8859
[43]	GENG J , WANG H Y , FAN J C ,et al. Deep supervised and contractive neural network for SAR image classification[J]. IEEE Transactions on Geoscience and Remote Sensing, 2017,55(4): 2442-2459. doi: 10.1109/TGRS.2016.2645226
[44]	SMOLENSKY P , . Information processing in dynamical systems:foundations of harmony theory[C]// Parallel Distributed Processing:Explorations in the Microstructure of Cognition,January 1-4,1986,Cambridge,USA. Cambridge:MIT Press, 1986.
[45]	WELLING M , ROSEN-ZVI M , HINTON G . Exponential family harmoniums with an application to information retrieval[C]// Advances in Neural Information Processing Systems 17,December 13-16,2004,Cambridge USA. Cambridge:MIT Press, 2005: 1481-1488.
[46]	HINTON G E . Training products of experts by minimizing contrastive divergence[J]. Neural Computation, 2002,14(8): 1771-1800. doi: 10.1162/089976602760128018 pmid: 12180402
[47]	ROUX N L , BENGIO Y . Representational power of restricted Boltzmann machines and deep belief networks[J]. Neural Computation, 2008,20(6): 1631-1649. doi: 10.1162/neco.2008.04-07-510 pmid: 18254699
[48]	徐丽坤, 刘晓东, 向小翠 . 基于深度信念网络的遥感影像识别与分类[J]. 地质科技情报, 2017,36(4): 244-249.
	XU L K , LIU X D , XIANG X C . Recognition and classification for remote sensing image based on depth belief network[J]. Geological Science and Technology Information, 2017,36(4): 244-249.
[49]	LIU Q , GAO Z Q , LIU B ,et al. Automated rule selection for opinion target extraction[J]. Knowledge-Based Systems, 2016(104): 74-88.
[50]	YOUNES L . On the convergence of markovian stochastic algorithms with rapidly decreasing ergodicity rates[J]. Stochastics and Stochastic Reports, 1999,65(3-4): 177-228. doi: 10.1080/17442509908834179
[51]	ALJARAH I , FARIS H , MIRJALILI S . Optimizing connection weights in neural networks using the whale optimization algorithm[J]. Soft Computing, 2018,22(1): 1-15. doi: 10.1007/s00500-016-2442-1
[52]	BENGIO Y , COURVILLE A , VINCENT P . Representation learning:a review and new perspectives[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2013,35(8): 1798-1828. doi: 10.1109/TPAMI.2013.50
[53]	杨建功, 汪西莉, 刘侍刚 . 融合谱-空域信息的 DBM 高光谱图像分类方法[J]. 西安电子科技大学学报, 2019,46(3): 109-115.
	YANG J G , WANG X L , LIU S G . Spectral-spatial classification of hyperspectral images using deep Boltzmann machines[J]. Journal of Xidian University, 2019,46(3): 109-115.
[54]	SALAKHUTDINOV R , LAROCHELLE H . Efficient learning of deep Boltzmann machines[J]. Journal of Machine Learning Research, 2010(9): 693-700. doi: 10.1016/j.neunet.2018.10.012 pmid: 30458316
[55]	SALAKHUTDINOV R , HINTON G . An efficient learning procedure for deep Boltzmann machines[J]. Neural Computation, 2012,24(8): 1967-2006.
[56]	SALAKHUTDINOV R , HINTON G . A better way to pretrain deep Boltzmann machines[C]// The 26th Annual Conference on Neural Information Processing Systems,December 3-6,2012,Lake Tahoe,Nevada,USA. Red Hook:Curran Associates Inc, 2012: 2447-2455.
[57]	CHO K , RAIKO T , ILIN A ,et al. A two-stage pretraining algorithm for deep Boltzmann machines[C]// 23rd International Conference on Artificial Neural Networks,Sep 10-Oct 13,2013,Techn Univ Sofia,Sofia,Bulgaria. Heidelberg:Springer, 2013: 106-113.
[58]	GOODFELLOW I , MIRZA M , COURVILLE A ,et al. Multi-prediction deep Boltzmann machines[C]// The 26th International Conference on Neural Information Processing Systems,December 5-10,2013,Lake Tahoe,Nevada,USA. Red Hook:Curran Associates Inc, 2013: 548-556.
[59]	BOURLARD H , KAMP Y . Auto-association by multilayer perceptrons and singular value decomposition[J]. Biological Cybernetics, 1988,59(4): 291-294. doi: 10.1007/bf00332918 pmid: 3196773
[60]	LEE H , GROSSE R , RANGANATH R ,et al. Convolutional deep belief networks for scalable unsupervised learning of hierarchical representations[C]// International Conference on Machine Learning,June 14-18,2009,Montreal,Canada. New York:ACM Press, 2009: 609-616.
[61]	HUANG G B , LEE H , LEARNED-MILLER E . Learning hierarchical representations for face verification with convolutional deep belief networks[C]// The IEEE Conference on Computer Vision and Pattern Recognition,Jun 16-21,2012,Washington,USA. Piscataway:IEEE Press, 2012: 2518-2525.
[62]	耿志强, 张怡康 . 一种基于胶质细胞链的改进深度信念网络模型[J]. 自动化学报, 2016,42(6): 943-952. doi: 10.16383/j.aas.2016.c150727
	GENG Z Q , ZHANG Y K . An improved deep belief network inspired by glia chains[J]. Acta Automatica Sinica, 2016,42(6): 943-952. doi: 10.16383/j.aas.2016.c150727
[63]	GOODFELLOW I J , POUGET-ABADIE J , MIRZA M ,et al. Generative adversarial nets[C]// Annual Conference on Neural Information Processing Systems,December 8-13,2014,Cambridge,USA. Cambridge:MIT Press, 2014: 2672-2680.
[64]	唐贤伦, 杜一铭, 刘雨微 ,等. 基于条件深度卷积生成对抗网络的图像识别方法[J]. 自动化学报, 2018,44(5): 855-864.
	TANG X L , DU Y M , LIU Y W ,et al. Image recognition with conditional deep convolutional generative adversarial networks[J]. Acta Automatica Sinica, 2018,44(5): 855-864.
[65]	SHERRINGTON C S . Observations on the scratch-reflex in the spinal dog[J]. The Journal of Physiology, 1906,34(1-2): 1-50. doi: 10.1113/jphysiol.1906.sp001139 pmid: 16992835
[66]	AKHTAR S W , REHMAN S , AKHTAR M ,et al. Improving the robustness of neural networks using k-support norm based adversarial training[J]. IEEE Access, 2016,4: 9501-9511. doi: 10.1109/ACCESS.2016.2643678
[67]	COOK J A , RANSTAM J . Overfitting[J]. British Journal of Surgery, 2016,103(13):1814. doi: 10.1002/bjs.10242 pmid: 27901285
[68]	ANTOL S , AGRAWAL A , LU J ,et al. VQA:visual question answering[C]// The 2015 IEEE International Conference on Computer Vision,December 7-13,2015,Santiago,Chile. Piscataway:IEEE Press, 2015: 2425-2433.
[69]	TUYTELAARS T , MIKOLAJCZYK K . Local invariant feature detectors:a survey[J]. Now Foundations and Trends, 2007,3(3): 177-280.
[70]	SQUARTINI S , PAOLINELLI S , PIAZZA F . Comparing different recurrent neural architectures on a specific task from vanishing gradient effect perspective[C]// 2006 IEEE International Conference on Networking,Sensing and Control,April 23-25,2006,FL,USA. Piscataway:IEEE Press, 2006: 380-385.
[71]	PASCANU R , MIKOLOV T , BENGIO Y . Understanding the exploding gradient problem[J]. Arxiv Preprint Arxiv, 2012.
[72]	HINTON G E , SRIVASTAVA N , KRIZHEVSKY A ,et al. Improving neural networks by preventing co-adaptation of feature detectors[J]. Computer Science, 2012,3(4): 212-223.
[73]	IOFFE S , SZEGEDY C . Batch normalization:accelerating deep network training by reducing internal covariate shift[C]// International Conference on Machine Learning,July 6-11,2015,Lile,France.[S.l.:s.n]. 2015: 448-456.
[74]	KRIZHEVSKY A , SUTSKEVER I , E.HINTON G . ImageNet classification with deep convolutional neural networks[C]// International Conference on Neural Information Processing Systems,December 3-6,2012,Lake Tahoe,Nevada. Red Hook:Curran Associates Inc, 2012: 1097-1105.
[75]	SIMONYAN K , ZISSERMAN A . Very deep convolutional networks for large-scale image recognition[C]// International Conference of Learning Representation,May 7-9,2015,San Diego,CA.arXiv:1409.1556v6[cs. CV], 2015.
[76]	HE K M , ZHANG X Y , REN S Q ,et al. Deep residual learning for image recognition[C]// IEEE Conference on Computer Vision and Pattern Recognition,June 27-30,2016,Las Vegas,Nevada. Los Alamitos:IEEE Computer Society, 2016: 770-778.
[77]	HE K M , ZHANG X Y , REN S Q ,et al. Identity mappings in deep residual networks[C]// 14th European Conference on Computer Vision,Octobet 8-16,2016,Amsterdam,Netherlands. Heidelberg:Springer, 2016: 630-645.
[78]	SZEGEDY C , LIU W , JIA Y Q ,et al. Going deeper with convolutions[C]// IEEE Conference on Computer Vision and Pattern Recognition,Juny 7-12,2015,Boston,MA,USA. Piscataway:IEEE Press, 2015: 1-9.
[79]	SZEGEDY C , VANHOUCKE V , IOFFE S ,et al. Rethinking the inception architecture for computer vision[C]// IEEE Conference on Computer Vision and Pattern Recognition,June 27-30,2016,Seattle,WA,USA. Piscataway:IEEE Press, 2016: 2818-2826.
[80]	HUANG G , LIU Z , MAATEN L V D ,et al. Densely connected convolutional networks[C]// IEEE Conference on Computer Vision and Pattern Recognition,July 21-26,2017,Honolulu,HI,USA. Piscataway:IEEE Press, 2017: 2261-2269.
[81]	IANDOLA F , HAN S , W.MOSKEWICZ M ,et al. SqueezeNet:AlexNet-level accuracy with 50x fewer parameters and <0.5MB model size[C]// International Conference on Learning Representations,April 24-26,2017,Toulon,France.arXiv:1602.07360v4[cs.CV]. 2016.
[82]	HOWARD A G , ZHU M L , CHEN B ,et al. MobileNets:efficient convolutional neural networks for mobile vision applications[J]. arXiv:1704.04861v1[cs.CV], 2017.
[83]	VASWANI A , SHAZEER N , PARMAR N ,et al. Attention is all you need[C]// 31st Conference on Neural Information Processing Systems,December 4-9,2017,Long Beach,CA,USA.[S.l.:s.n]. 2017.
[84]	WANG F , JIANG M Q , QIAN C ,et al. Residual attention network for image classification[C]// IEEE Conference on Computer Vision and Pattern Recognition,July 21-26,2017,Honolulu,HI,USA. Piscataway:IEEE Press, 2017: 6450-6458.
[85]	HU J , SHEN L , SUN G . Squeeze-and-excitation networks[C]// IEEE Conference on Computer Vision and Pattern Recognition,June 18-23,2018,New York,USA. Piscataway:IEEE Press, 2018: 7132-7141.
[86]	SABOUR S , FROSST N , E HINTON G . Dynamic routing between capsules[C]// 31st Conference on Neural Information Processing Systems,December 4-9,2017,Long Beach,CA,USA.arXiv:1710.09829v2[cs.CV]. 2017.

深度学习模型	错误率
KNN^[13]	5.0%
MLP^[18]	2.45%
传统自编码器^[41]	1.78%
堆叠去噪自编码器^[38]	1.28%
DBN^[25]	1.25%
LeNet-5^[18]	0.8%

CNN模型	Top-1错误率（val）	Top-5错误率（val）	Top-5错误率（test）
AlexNet^[74]	36.7%	15.4%	15.3%
VGGNet^[75]	—	8.43%	7.32%
GoogleNet^[78]	—	7.89%	6.66%
BN-Inception^[73]	21.99%	5.82%	—
ResNet-152^[76]	19.38%	4.49%	3.57%
Inception-V3^[79]	18.77%	4.2%	—
DensNet-264^[80]	20.80%	5.29%	—
Attention-92^[84]	19.5%	4.8%	—
1.0 MobileNet-224^[82]	29.4%	—	—
SENet-154^[85]	18.68%	4.47%	2.25%

CNN模型	FLOPS （每秒浮点运算次数）	参数数量/个
LeNet-5^[18]	4×10⁵	6×10⁴
AlexNet^[74]	7×10⁹	6×10⁷
VGGNet-16^[75]	1.5×10¹¹	1.38×10⁸
GoogleNet^[78]	1.5×10¹⁰	5×10⁶
ResNet-50^[76]	3.86×10⁹	2.5×10⁷
Attention-92^[84]	1.04×10¹⁰	5.13×10⁷
MobileNet^[82]	5.69×10⁸	4.2×10⁶
SENet-50^[85]	3.87×10⁹	2.75×10⁷