Hybrid Neuro-Fuzzy Model with Immune Training for Recognition of Objects in an Image

Mykola Korablyov, Natalia Axak, Oleksandr Fomichov, Andrii Chuprina
2020 International Conference on Information Control Systems & Technologies  
Modern systems for image processing and analysis are characterized by the active use of artificial neural networks, for training of which, as a rule, gradient methods are used, but their main limitation of the implementation is high computational cost. The use of the principles of hybridization of neural networks, fuzzy logic and evolutionary algorithms allows you to create new types of models that have a higher recognition quality while reducing the computational cost of training. A hybrid
more » ... o-fuzzy recognition model is proposed, which consists of two modules: a convolutional module (CNN) and a neuro-fuzzy classifier module (NFC) built on the basis of a modified ANFIS network. The CNN module, which is trained by the method of back propagation error, acts as a kind of expert system for the NFC module. It is proposed to perform NFC training based on the use of artificial immune systems by presenting all adjustable parameters in the form of a structured adaptive multiantibody, and consists in adjusting the NFC parameters and structure. Experimental studies have been carried out on test samples, confirming the effectiveness of the proposed model for recognizing objects in an image.
dblp:conf/icst2/KorablyovAFC20 fatcat:acwkcxtlvfaqfgestvbqrhznu4