NEURAL NETWORK ALGORITHMS IN THE PROBLEMS OF AUTOMATIC RECOGNITION AND TRACKING OF OBJECTS FOR OPTICAL-ELECTRONIC SYSTEMS EMBEDED PLATFORMS
Abstract
The goal of this paper is to improve the efficiency of automatic tracking and recognition sys-tems of local objects in difficult observation conditions. In automatic object recognition tasks, it is often difficult to predict object’s angle of observation, and therefore impossible to prepare re-quired etalons. In such conditions the best performance is shown by the algorithms based on deep convolutional neural networks (CNN). The novelty of this work is in the development of original object recognition algorithm using CNN. This algorithm belongs to the group of supervised ma-chine learning algorithm. The main feature of algorithms of this group is their good scalability on new object image samples that not represented in a training set. Typical CNN development pro-cess can be divided in three main stages: researching and architecture selection, training and CNN application. On first stage the research of application field is performed, typical objects images analysis is conducted, typical object class amount is determined, training set preparation, and CNN architecture selection. The result of the first stage of development is a defined CNN ar-chitecture and prepared training set. On the next stage iterative CNN training process is per-formed CNN architecture correction possibility depending on training metrics analysis. The train-ing is conducted in laboratory conditions on powerful workstations. On CNN application stage the recognition procedure by train CNN is performed, also known as forward propagation. This stage is being performed on embedded systems. The key difference of developed CNN from wide group of similar known algorithms is its fundamental orientation usage in embedded systems, which are have strict restrictions on their mass, dimensions, energy consumption, etc. This is achieved by using of integer binary arithmetic, which allows its efficient FPGA implementation. The efficien-cy evaluation of developed CNN has been conducted with the help of complex mathematical model of hardware and software complex of electro-optical systems (EOS). The developed algorithm was tested as part of special purpose software of unmanned aerial vehicle EOS and demonstrated high practical efficiency.
References
2. Han S., Mao H., Dally W.J. Deep compression: Compressing deep neural networks with prun-ing, trained quantization and huffman coding.
3. Courbariaux M., Bengio Y., David J.P. Binaryconnect: Training deep neural networks with binaryweights during propagations // Advances in neural information processing systems. – 2015. – P. 3123-3131.
4. Ioffe S., Szegedy C. Batch normalization: Accelerating deep network training by reducing in-ternal covariate shift // arXiv preprint arXiv:1502.03167. 2015.
5. Courbariaux M. et al. Binarized neural networks: Training deep neural networks with weights and activations constrained to+ 1 or-1 // arXiv preprint arXiv:1602.02830. 2016.
6. Hinton Geoffrey. Neural networks for machine learning. Coursera. Video lectures. – 2012.
7. Rastegari M. et al. Xnor-net: Imagenet classification using binary convolutional neural net-works // European Conference on Computer Vision. – Springer, Cham, 2016. – P. 525-542.
8. Jia Y. et al. Caffe: convolutional architecture for fast feature embedding // Proceedings of the 22nd ACM International Conference on Multimedia. – 2014. – P. 675-678.
9. Kingma D., Ba J. Adam. A method for stochastic optimization // International Conference on Multimedia. – 2014. – P. 675-678.
10. Тупиков В.А., Павлова В.А., Бондаренко В.А., Александров В.А. Способ автоматического обнаружения объектов на морской поверхности в видимом диапазоне // Известия Тул-ГУ. Технические науки. – 2016. – № 11-3. – С. 105-121.
11. Gonzalez R.C., Woods R.E. Digital Image Processing. – 2nd ed. – Prentice-Hall, 2002. – 793 p.
12. Sauvola J., Pietikainen M. Adaptive document image binarization // Pattern Recognition. – 2000. – No. 33. – P. 225-236.
13. Shafait F., Keysers D., Breuel T.M. Efficient implementation of local adaptive thresholding techniques using integral images // Proc. SPIE vol. 6815, Document Recognition and Retrieval XV, 2008.
14. Viola P., Jones M.J. Robust real-time face detection // Int. Journal of Computer Vision. – 2004. – Vol. 57 (2). – P. 137-154.
15. Crow F. Summed-area tables for texture mapping // In Proceedings of SIGGRAPH. – 1984. – Vol. 18 (3). – P. 207-212.
16. Баклицкий В.К. Корреляционно-экстремальные методы навигации и наведения. – Тверь: ТО «Книжный клуб», 2009. – 360 с.
17. LeCun Y., Boser B., Denker J.S., Henderson D., Howard R.E., Hubbard W. and Jackel L.D. Backpropagation Applied to Handwritten Zip Code Recognition // Neural Computation. – 1989. – No. 1 (4). – P. 541-551.
18. Goodfellow I., Bengio Y., Courville A. Deep Learning. – MIT Press, 2016. – 781 p.
19. Deng L. and Deep D.Yu. Learning: Methods and Applications // Foundations and Trends in Signal Processing. – 2013. – Vol. 7, No. 3–4. –P. 197-387.
20. Ioffe, Sergey, and Christian Szegedy. Batch normalization: Accelerating deep network training by reducing internal covariate shift // arXiv preprint. – 2015. arXiv:1502.03167.
