Peleshchak I. Multispectral image recognition system based on oscillatory neural networks

In the dissertation work on the basis of the methodology of system analysis the system of recognition and encryption of multispectral images on the basis of oscillatory neural networks is developed. The structure and properties of artificial neural networks with linear and nonlinear oscillatory neurons has been studied by the methods of system analysis; a method of compression of input images in a neural network due to diagonalization of a matrix of weight synaptic connections has been developed; a method for encrypting information with a constantly changing asymmetric key for each new input image has been developed, which is based on the synthesis of a diagonalized neural network and the AES (Advanced Encryption Standard (Rijndael)) algorithm; a method of information encryption based on nonlinear oscillatory neurons with a chain and ring topology has been developed; a method for recognizing multispectral images by an oscillatory neural network based on information resonance has been developed; a model for optimizing the size (number of neurons, number of synaptic connections) within a nonlinear generalized error has been developed; in Python developed a program for recognition of multispectral images based on information resonance using a three-layer oscillatory neural network; conducted a computer experiment to calculate the parameter of the complexity of learning a three-layer neural network (the number of operations performed by the neural network) and its optimal size.