Performance Enhanced HBONet CNN Approach for Embedded Systems

Abstract

In recent years, the usage areas of convolutional neural networks (CNN) have increased remarkably. It is widely used on many platforms, from workstations to embedded devices. However, each CNN model uses a different amount of memory, processor, storage and has different object recognition accuracy rates. CNNs to be used in embedded systems have some difficulties such as being less costly, consuming less resources and achieving higher accuracy. One of the CNN models that best overcomes these difficulties is the HBONet model. However, this model does not perform well enough in embedded systems. In this study, it is aimed to increase the performance of the HBONet model for embedded systems. For this purpose, the A-HBONet model, which is based on the HBONet model, is proposed. As a result of the experiments performed, the accuracy of the proposed model was increased by 3% compared to the HBONet model, while the memory and storage unit usage was reduced by approximately 80%. These results show that the proposed model works more effectively and efficiently in embedded devices.

Keywords

• Embedded systems
• convolutional neural networks
• A-HBONet
• CIFAR-10

Gömülü Sistemler İçin Performansı Arttırılmış HBONet CNN Yaklaşımı

Abstract

Son yıllarda, evrişimli sinir ağlarının (CNN) kullanım alanları dikkate değer bir şekilde artmıştır. İş istasyonlarından gömülü cihazlara varıncaya kadar birçok platformda yaygın olarak kullanılmaktadır. Bununla birlikte, her CNN modeli farklı miktarda hafıza, işlemci, depolama birimi kullanmaktadır ve nesne tanımada farklı doğruluk oranlarına sahiptir. Gömülü sistemlerde kullanılacak CNN’lerin daha az maliyetli olması, daha az kaynak tüketmesi ve daha fazla doğruluk oranını başarması gibi bazı zorlukları vardır. Bu zorlukların en iyi üstesinden gelen CNN modellerinden biri de HBONet modelidir. Ancak, bu model gömülü sistemlerde yeterince iyi performans sağlamamaktadır. Bu çalışmada, gömülü sistemler için kullanılan HBONet modelinin kaynak tüketimi ve doğruluk gibi performans metriklerinin daha da iyileştirilmesi amaçlanmıştır. Bu amaçla, HBONet modelini temel alan bir model olan A-HBONet modeli önerilmiştir. CIFAR-10 veri seti kullanılarak gerçekleştirilen deneyler sonucunda, önerilen modelin doğruluğu HBONet modeline göre %3 arttırılırken hafıza ve depolama birimi kullanımı da yaklaşık olarak %80 oranında azaltılmıştır. Bu sonuçlar, önerilen modelin gömülü cihazlarda daha etkin ve verimli çalıştığı göstermektedir.

Keywords

• Gömülü sistemler
• evrişimli sinir ağları
• A-HBONet
• CIFAR-10

References

1. DeVries, T., & Taylor, G. W. (2017). Improved Regularization of Convolutional Neural Networks with Cutout. Retrieved from http://arxiv.org/abs/1708.04552
2. Han, S., Pool, J., Tran, J., & Dally, W. J. (2015). Learning both weights and connections for efficient neural networks. Advances in Neural Information Processing Systems, 2015-Janua, 1135–1143.
3. Howard, A. G., Zhu, M., Chen, B., Kalenichenko, D., Wang, W., Weyand, T., … Adam, H. (2017). MobileNets: Efficient convolutional neural networks for mobile vision applications. ArXiv.
4. Koonce, B. (2021). MobileNet v1. Convolutional Neural Networks with Swift for Tensorflow, 87–97. https://doi.org/10.1007/978-1-4842-6168-2_8
5. Krizhevsky, A. (2009). CIFAR10 Dataset. Retrieved from https://www.cs.toronto.edu/ kriz/cifar.html
6. Krizhevsky, A., Sutskever, I., & Hinton, G. E. (2012). ImageNet classification with deep convolutional neural networks. Proceedings of the 25th International Conference on Neural Information Processing Systems, 1, 1097–1105. https://doi.org/10.1145/3065386
7. Li, D., Zhou, A., & Yao, A. (2019). HBONet: Harmonious bottleneck on two orthogonal dimensions. Proceedings of the IEEE International Conference on Computer Vision, 2019-Octob, 3315–3324. https://doi.org/10.1109/ICCV.2019.00341
8. Loshchilov, I., & Hutter, F. (2017). SGDR: Stochastic gradient descent with warm restarts. 5th International Conference on Learning Representations, ICLR 2017 - Conference Track Proceedings, 1–16.

9. Ma, N., Zhang, X., Zheng, H. T., & Sun, J. (2018). Shufflenet V2: Practical guidelines for efficient cnn architecture design. Lecture Notes in Computer Science (Including Subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), 11218 LNCS, 122–138. https://doi.org/10.1007/978-3-030-01264-9_8
10. Rinner, B., & Wolf, W. (2008). An introduction to distributed smart cameras. Proceedings of the IEEE, 96(10), 1565–1575. https://doi.org/10.1109/JPROC.2008.928742
11. Sandler, M., Howard, A., Zhu, M., Zhmoginov, A., & Chen, L. C. (2018a). MobileNetV2: Inverted residuals and linear bottlenecks. ArXiv, 4510–4520.
12. Sandler, M., Howard, A., Zhu, M., Zhmoginov, A., & Chen, L. C. (2018b). MobileNetV2: Inverted Residuals and Linear Bottlenecks. Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition, 4510–4520. https://doi.org/10.1109/CVPR.2018.00474
13. Shawahna, A., Sait, S. M., & El-Maleh, A. (2019). FPGA-Based accelerators of deep learning networks for learning and classification: A review. IEEE Access, 7, 7823–7859. https://doi.org/10.1109/ACCESS.2018.2890150
14. Simonyan, K., & Zisserman, A. (2015). Very deep convolutional networks for large-scale image recognition. 3rd International Conference on Learning Representations, ICLR 2015 - Conference Track Proceedings, 1–14.
15. Weibin Liu, Chao Zhang, & Baozong Yuan. (2002). AVR theory, techniques and application. (69775003), 1163–1166. https://doi.org/10.1109/icosp.2000.891751
16. Zhang, X., Zhou, X., Lin, M., & Sun, J. (2018). ShuffleNet: An Extremely Efficient Convolutional Neural Network for Mobile Devices. Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition, 6848–6856. https://doi.org/10.1109/CVPR.2018.00716