Fuzzy Classification on the Base of Convolutional Neural Networks

The paper deals with the algorithm of object classification based on the method of fuzzy logic and the application of artificial convolutional neural networks. Every object can be characterized by a set of data presented in the numerical form and in the form of images (photographs in different parts of the light spectrum). In this case, one object can be matched with a few images associated with it; they can be received by different methods and from different sources. In the algorithm, this generalized totality of images is recognized by convolutional neural networks. A separate neural network is formed for every channel of data receiving. Then, the network outputs are combined for processing in the system of classification on the basis of fuzzy logic output. The normalized outputs of convolutional neural networks are used as values of a membership function to terms of outputs variables when a fuzzy classifier works. For the first adjustment of the convolution neural network hyperparameters, the gradient method is applied. The algorithm is realized in Python language with the use of Keras deep learning library and Tensor Flow library of parallel computation with CUDA technology from NVIDIA company. This paper presents the results of practical application of the developed neuro-fuzzy classifier to forecast the problem of working time losses.