Simulating primary visual cortex to improve robustness of CNN neural network structures

ZHANG Lijuan; HU Mengda; ZHANG Ziwei; JIANG Yutong; LI Dongming

doi:10.37188/OPE.20233115.2287

您当前的位置：

首页 >

文章列表页 >

Simulating primary visual cortex to improve robustness of CNN neural network structures

Information Sciences | 更新时间：2023-08-28

- Simulating primary visual cortex to improve robustness of CNN neural network structures
- Optics and Precision Engineering Vol. 31, Issue 15, Pages: 2287-2294(2023)
- 作者机构：
  
  1.无锡学院物联网工程学院，江苏无锡 214105
  2.长春工业大学计算机科学与工程学院，吉林长春 130012
  3.中国北方车辆研究所，北京100072
- 作者简介：
  
  E-mail： LDM0214@163.com
- 基金信息：
- DOI：10.37188/OPE.20233115.2287
  CLC：
扫描看全文
ZHANG Lijuan, HU Mengda, ZHANG Ziwei, et al. Simulating primary visual cortex to improve robustness of CNN neural network structures. [J]. Optics and Precision Engineering 31(15):2287-2294(2023)
DOI：

ZHANG Lijuan, HU Mengda, ZHANG Ziwei, et al. Simulating primary visual cortex to improve robustness of CNN neural network structures. [J]. Optics and Precision Engineering 31(15):2287-2294(2023) DOI： 10.37188/OPE.20233115.2287.

摘要

针对卷积网络模型的稳定性能较差，对抗训练方法会使得网络结构过于复杂并占用大量运算资源的问题，提出了一种基于人体视觉神经系统生物特征的卷积神经网络模型改进方法（VVNet）。在卷积神经网络的基础上，融合人体视觉的结构特征，在不增加网络层数或保持准确率不变的情况下，提高神经网络面对噪声干扰的稳定性。在数据集Cifar10上对3种不同神经网络模型（VVNet，VOneNet以及原网络模型）进行测试。实验结果表明，使用VVNet网络模型、VOneNet网络模型和原始的网络模型DenseNet121对四类图像（噪声图像、模糊图像、遮挡图像和饱和曝光图像）的分类准确率进行对比，验证了提出的VVNet网络结构对不同类型图像的分类准确率几乎不变，在使用对抗样本情况下，VVNet网络结构的图像分类准确率提高了约10%。与深度学习网络相比，基于人体视觉系统结构的网络能够在保持准确率的同时有效地提高神经网络的稳定性，并具有可移植性。

Abstract

The robustness of convolutional neural network （CNN） models is usually improved by deepening the number of network layers to ensure the accuracy of the results. However， increasing the number of network layers will make the network more complex and occupy more space. This paper proposes an improved CNN modeling method based on human visual features. Through the CNN， the structural features of human vision are fused to improve the robustness of the network against noise without increasing the number of layers or affecting the original accuracy of the model. The experimental results on the Cifar10 dataset show that the classification accuracy of the image inserted into the proposed VVNet is almost the same as that of the original network， and the classification accuracy is improved by approximately 10% in the case of image destruction. Compared with the original deep learning network， the network based on human visual system structure can effectively enhance the robustness of the network while maintaining the original accuracy.

关键词

计算机视觉机器学习图像识别视觉皮层

Keywords

computer visionmachine learningimage recognitionvisual cortex

references

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: 10.1109/5.726791http://dx.doi.org/10.1109/5.726791

KRIZHEVSKY A， SUTSKEVER I， HINTON G E. ImageNet classification with deep convolutional neural networks［J］. Communications of the ACM， 2017， 60（6）： 84-90. doi: 10.1145/3065386http://dx.doi.org/10.1145/3065386

HE K M， ZHANG X Y， REN S Q， et al. Deep residual learning for image recognition［C］. 2016 IEEE Conference on Computer Vision and Pattern Recognition （CVPR）.2730，2016， Las Vegas， NV， USA. IEEE， 2016： 770-778. doi: 10.1109/cvpr.2016.90http://dx.doi.org/10.1109/cvpr.2016.90

SZEGEDY C， VANHOUCKE V， IOFFE S， et al. Rethinking the inception architecture for computer vision［C］. 2016 IEEE Conference on Computer Vision and Pattern Recognition （CVPR）.2730，2016， Las Vegas， NV， USA. IEEE， 2016： 2818-2826. doi: 10.1109/cvpr.2016.308http://dx.doi.org/10.1109/cvpr.2016.308

HUANG G， LIU Z， VAN DER MAATEN L， et al. Densely connected convolutional networks［C］. 2017 IEEE Conference on Computer Vision and Pattern Recognition （CVPR）.2126，2017， Honolulu， HI， USA. IEEE， 2017： 2261-2269. doi: 10.1109/cvpr.2017.243http://dx.doi.org/10.1109/cvpr.2017.243

SZEGEDY C， ZAREMBA W， SUTSKEVER I， et al. Intriguing properties of neural networks［EB/OL］. 2013： arXiv： 1312.6199. https：//arxiv.org/abs/1312.6199https://arxiv.org/abs/1312.6199.

ILYAS A， SANTURKAR S， TSIPRAS D， et al. Adversarial examples are not bugs， they are features［EB/OL］. 2019： arXiv： 1905.02175. https：//arxiv.org/abs/1905.02175https://arxiv.org/abs/1905.02175. doi: 10.23915/distill.00019http://dx.doi.org/10.23915/distill.00019

林点，潘理，易平. 面向图像识别的卷积神经网络稳定性研究进展［J］. 网络与信息安全学报， 2022， 8（3）：111-122. doi: 10.11959/j.issn.2096-109x.2022037http://dx.doi.org/10.11959/j.issn.2096-109x.2022037

LIN D， PAN L， YI P. Research on the robustness of convolutional neural networks in image recognition［J］. Chinese Journal of Network and Information Security， 2022， 8（3）：111-122.（in Chinese）. doi: 10.11959/j.issn.2096-109x.2022037http://dx.doi.org/10.11959/j.issn.2096-109x.2022037

DZIUGAITE G K， GHAHRAMANI Z， ROY D M. A study of the effect of JPG compression on adversarial images［EB/OL］. 2016： arXiv： 1608.00853. https：//arxiv.org/abs/1608.00853https://arxiv.org/abs/1608.00853.

VINCENT P， LAROCHELLE H， BENGIO Y， et al. Extracting and composing robust features with denoising autoencoders［C］. Proceedings of the 25th international conference on Machine learning. 59，2008， Helsinki， Finland. New York： ACM， 2008： 1096-1103. doi: 10.1145/1390156.1390294http://dx.doi.org/10.1145/1390156.1390294

XU W， EVANS D， QI Y. Feature squeezing： detecting adversarial examples in deep neural networks［EB/OL］. 2017： arXiv： 1704.01155. https：//arxiv.org/abs/1704.01155https://arxiv.org/abs/1704.01155. doi: 10.14722/ndss.2018.23198http://dx.doi.org/10.14722/ndss.2018.23198

TSIPRAS D， SANTURKAR S， ENGSTROM L， et al. Robustness may be at odds with accuracy［EB/OL］. 2018： arXiv： 1805.12152. https：//arxiv.org/abs/1805.12152https://arxiv.org/abs/1805.12152.

XIE C H， TAN M X， GONG B Q， et al. Adversarial examples improve image recognition［C］. 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition （CVPR）.1319，2020， Seattle， WA， USA. IEEE， 2020： 816-825. doi: 10.1109/cvpr42600.2020.00090http://dx.doi.org/10.1109/cvpr42600.2020.00090

CARLINI N， WAGNER D. Adversarial examples are not easily detected： bypassing ten detection methods［C］. Proceedings of the 10th ACM Workshop on Artificial Intelligence and Security. 3 November 2017， Dallas， Texas， USA. New York： ACM， 2017： 3-14. doi: 10.1145/3128572.3140444http://dx.doi.org/10.1145/3128572.3140444

ULLMAN S， ASSIF L， FETAYA E， et al. Atoms of recognition in human and computer vision［J］. Proceedings of the National Academy of Sciences of the United States of America， 2016， 113（10）： 2744-2749. doi: 10.1073/pnas.1513198113http://dx.doi.org/10.1073/pnas.1513198113

HUBEL D H， WIESEL T N. Receptive fields， binocular interaction and functional architecture in the cat’s visual cortex［J］. The Journal of Physiology， 1962， 160（1）： 106-154. doi: 10.1113/jphysiol.1962.sp006837http://dx.doi.org/10.1113/jphysiol.1962.sp006837

ZADOR A M. A critique of pure learning and what artificial neural networks can learn from animal brains［J］. Nature Communications， 2019， 10： 3770. doi: 10.1038/s41467-019-11786-6http://dx.doi.org/10.1038/s41467-019-11786-6

MARBLESTONE A H， WAYNE G， KORDING K P. Toward an integration of deep learning and neuroscience［J］. Frontiers in Computational Neuroscience， 2016， 10： 94. doi: 10.3389/fncom.2016.00094http://dx.doi.org/10.3389/fncom.2016.00094

NAYEBI A， GANGULI S. Biologically inspired protection of deep networks from adversarial attacks［EB/OL］. 2017： arXiv： 1703.09202. https：//arxiv.org/abs/1703.09202https://arxiv.org/abs/1703.09202.

LINDSAY G W， MILLER K D. How biological attention mechanisms improve task performance in a large-scale visual system model［J］. eLife， 2018， 7： 38105. doi: 10.7554/elife.38105http://dx.doi.org/10.7554/elife.38105

HASANI H， SOLEYMANI M， AGHAJAN H. Surround modulation： a bio-inspired connectivity structure for convolutional neural networks［J］. Advances in Neural Information Processing Systems， 2019.

REDDY M V， BANBURSKI A， PANT N， et al. Biologically inspired mechanisms for adversarial robustness［EB/OL］. 2020： arXiv： 2006.16427. https：//arxiv.org/abs/2006.16427https://arxiv.org/abs/2006.16427.

KIM E， REGO J， WATKINS Y， et al. Modeling biological immunity to adversarial examples［C］. 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition （CVPR）.1319，2020， Seattle， WA， USA. IEEE， 2020： 4665-4674. doi: 10.1109/cvpr42600.2020.00472http://dx.doi.org/10.1109/cvpr42600.2020.00472

DAPELLO J， MARQUES T， SCHRIMPF M， et al. Simulating a primary visual cortex at the front of CNNs improves robustness to image perturbations［J］. Advances in Neural Information Processing Systems， 2020， 33： 13073-13087.

JONES J P， PALMER L A. An evaluation of the two-dimensional Gabor filter model of simple receptive fields in cat striate cortex［J］. Journal of Neurophysiology， 1987， 58（6）： 1233-1258. doi: 10.1152/jn.1987.58.6.1233http://dx.doi.org/10.1152/jn.1987.58.6.1233

ADELSON E H， BERGEN J R. Spatiotemporal energy models for the perception of motion［J］. Journal of the Optical Society of America A， 1985， 2（2）： 284. doi: 10.1364/josaa.2.000284http://dx.doi.org/10.1364/josaa.2.000284

VINTCH B， MOVSHON J A， SIMONCELLI E P. A convolutional subunit model for neuronal responses in macaque V1［J］. The Journal of Neuroscience， 2015， 35（44）： 14829-14841. doi: 10.1523/jneurosci.2815-13.2015http://dx.doi.org/10.1523/jneurosci.2815-13.2015

CADENA S A， DENFIELD G H， WALKER E Y， et al. Deep convolutional models improve predictions of macaque V1 responses to natural images［J］. PLoS Computational Biology， 2019， 15（4）： e1006897. doi: 10.1371/journal.pcbi.1006897http://dx.doi.org/10.1371/journal.pcbi.1006897

SCHRIMPF M， KUBILIUS J， HONG H， et al. Brain-Score： which Artificial Neural Network for Object Recognition is most Brain-Like？［J］. bioRxiv， 2018， DOI： 10.1101/407007http://dx.doi.org/10.1101/407007.

KUBILIUS J， SCHRIMPF M， HONG H， et al. Brain-like object recognition with high-performing shallow recurrent ANNs［EB/OL］. 2019： arXiv： 1909.06161. https：//arxiv.org/abs/1909.06161https://arxiv.org/abs/1909.06161.

EL-SHAMAYLEH Y， KUMBHANI R D， DHRUV N T， et al. Visual response properties of V1 neurons projecting to V2 in macaque［J］. The Journal of Neuroscience， 2013， 33（42）： 16594-16605. doi: 10.1523/jneurosci.2753-13.2013http://dx.doi.org/10.1523/jneurosci.2753-13.2013

SOFTKY W R， KOCH C. The highly irregular firing of cortical cells is inconsistent with temporal integration of random EPSPs［J］. The Journal of Neuroscience， 1993， 13（1）： 334-350. doi: 10.1523/jneurosci.13-01-00334.1993http://dx.doi.org/10.1523/jneurosci.13-01-00334.1993

LINDSEY J， OCKO S A， GANGULI S， et al. A unified theory of early visual representations from retina to cortex through anatomically constrained deep CNNs［EB/OL］. 2019： arXiv： 1901.00945. https：//arxiv.org/abs/1901.00945https://arxiv.org/abs/1901.00945.

SIMONYAN K， ZISSERMAN A. Very deep convolutional networks for large-scale image recognition［EB/OL］. 2014： arXiv： 1409.1556. https：//arxiv.org/abs/1409.1556https://arxiv.org/abs/1409.1556.

Views

下载量

CSCD

Alert me when the article has been cited

提交

Tools

Publicity Resources

Uniform defocus blind deblurring based on deeper feature-based wiener deconvolution

Partial optimal transport-based domain adaptation for hyperspectral image classification

Multi-stage frame alignment video super- resolution network

Semi-supervised instance object detection method based on SVD co-training

Road traffic sign recognition algorithm based on improved YOLOv4

Related Author

No data

Related Institution

School of Software Engineering， Southeast University

School of Mechanical Engineering， Southeast University， Nanjin

Wuxi shangshi-finevision Technology Co.， Ltd

College of Computer Science and Technology， Jilin University

Suzhou Institute of Biomedical Engineering and Technology， Chinese Academy of Sciences

Address：No.3888 Dong Nanhu Road, Changchun, Jilin, China Postal code：130033
Tel：0431-86176855 Email：gxjmgc@ciomp.ac.cn
Technical support is provided by Beijing Founder electronics co., LTD 吉ICP备11002662号-17 京公网安备11010802024621
It is recommended to read the content of this site in Chrome&IE9+. Please switch to extreme mode in browser 360.
Cookies We use cookies to help provide and enhance our service and tailor content. By continuing, you agree to the use of cookies.

⁰