The Method of Intelligent Image Processing Based on a Three-Channel Purely Convolutional Neural Network

Abstract

In the paper a method of intelligent image processing have been developed. This method based on a three channel purely convolutional neural network. The method consists of neural network model, a criterion to evaluate the effectiveness of the proposed model, a method for neural network learning in batch mode and a corresponding learning algorithm. This algorithm is intended for implementation on GPU by means of CUDA technology. The created model of neural network does not require the determination of the number of planes in hidden layers. This feature of the model simplifies its parametric identification "in large" and provides the use of three planes in the input layer. This simplifies the work with RGB images. The proposed method of intelligent image processing can be used in various intelligent systems of medical diagnostics.