Novoseltsev I. Methods and means for recognizing the changes of the object properties through images based on artificial neural networks

The actual scientific and practical problem of developing neural network methods for the recognition and classification of changes of the image properties based on the received theoretical and experimental researches is solved. In contrast to the existing methods, they allow reducing the recognition error and increasing the accuracy of the image classification in the conditions of a priori and current uncertainty and the presence of interference. The object of the study is the process of recognizing changes in the properties of objects by the image. A new method for controlling the change of an object size using a reference with the introduction of the properties of the transformation of object similarity is developed, which allows increasing the accuracy of measurements. In order to increase the robustness of BP parameters estimates, a training procedure for BP is developed, it is a matrix version of the Kaczmarz’s procedure (Widrow-Hoff), its proposed modification contains a zone of insensitivity; the procedures for adjusting this zone are considered. The neural network image classification method, based on convolutional neural networks, has been further developed using various activation functions and robust learning of network parameters in different layers. The neural network method for recognizing the change of image parameters on the example of melanoma using the PNN and CNN networks has received further development on the basis of which a neural network method for controlling the change in the size of each formation and recognition of the heterogeneity of the color of skin changes has been developed. Software-implemented neural network system for determining the heterogeneity of coloring, consisting of a group of networks PNN and CNN, which allows to improve the accuracy of diagnosis, reduce the time of training, reduce the risk of excessive training and makes it easier for the doctor to diagnose.