N.M. Novikova - Dr. Sc. (Eng.), Professor, Voronezh State University. E-mail: novnelly839@gmail.com V.M. Dudenkov - Post-graduate Student, Voronezh State University. E-mail: vldud@mail.ru

In this article a problem of big size images recognition is considered under conditions of minimum volume of a training set. The following solution of the problem is provided: union of convolutional neural networks and a fuzzy logic network. Image segmentation via agglomerative clustering algorithm is used for additional optimization. After segmentation some fragments of the image are sent to inputs of different convolutional neural networks. Outputs of these convolutional neural networks in its turn are sent to inputs of the fuzzy logic network. In such architecture of a neural network the convolutional neural networks are responsible for preliminary image processing, but the fuzzy neural network are responsible for a final decision about classification accessory. Thus, the developed neural network shown a high recognition ability and low requirements for computing resources.

 

1. Spicin V.G. Algoritm obnaruzhenija lic na osnove svertochnojj nejjronnojj seti // Nejjrokompjutery: razrabotka, primenenie. 2013. № 10. S. 48−52.
2. Gonsales R., Vuds R. Cifrovaja obrabotka izobrazhenijj. M.: Tekhnosfera. 2005. 1007 s.
3. Novikova N.M., Budko V.N. Intellektualnye interfejjsy. Voronezh: IPC VGU. 2011. 308 s.
4. LeCun Y. and Bengio Y. Convolutional Networks for Images, Speech and Time-Series // The Handbook of Brain Theory and Neural Networks. Arbib M.A. (Eds). MITPress. 1995.
5. KHajjkin S. Nejjronnye seti: polnyjj kurs. Izd. 2‑e, ispravl. M.: Izdatelskijj dom «Viljams». 2006. 1104 s.
6. LeCun Y., Bottou L., Orr G. and Muller K. Efficient BackProp // Neural Networks: Tricks of the trade. Springer. 1998.
7. Kruglov V.V., Dli M.I., Golunov R.JU. Nechetkaja logika i iskusstvennye nejjronnye seti. M.: FIZMATLIT. 2001. 201 s.