INFORMATION-EXTREME HIERARCHICAL MACHINE LEARNING OF THE HAND BRUSH PROSTHESIS CONTROL SYSTEM WITH A NON-INVASIVE BIO SIGNAL READING SYSTEM

Abstract

Context. The actual problem of the information synthesis of learning and control systems for the prosthesis of the hand with a non-invasive system for reading biosignals has been solved.Objective. The goal of the work is to increase the functional efficiency of the control system for the prosthesis of the hand with a non-invasive system for reading biosignals based on machine learning, which allows the system to operate in the operating mode to recognize the cognitive commands of the user of the prosthesis with high reliability and efficiency.Method. Within the framework of informational and extreme intellectual technology (IEI technology) of data analysis based on maximizing the informational ability of a recognition system in machine learning, a method of informational synthesis of an intelligent control system for a prosthetic hand with a non-invasive biosignal reading system is proposed. In contrast to the existing methods of data mining, the method of information-extremal machine learning was developed as part of a functional approach to modeling the cognitive processes inherent in humans in the formation and adoption of classification decisions. This approach makes it possible to endow the prosthesis management system with adaptability properties to arbitrary initial conditions for the formation of cognitive teams and retraining while expanding the vocabulary of signs and the alphabet of recognition classes. In addition, the decision rules based on the geometric parameters of hyperspherical containers of recognition classes obtained during machine learning are almost invariant to the multidimensionality of the recognition feature space. Based on the proposed category model, a machine learning algorithm has been developed with optimization of the hierarchical data structure. At the same time, the influence on the functional efficiency of machine learning of data structures constructed in the form of dichotomous and decursion trees was studied. As a criterion for optimizing machine learning parameters, a modification of the informational Kullback measure is used, which is a functional of the accuracy characteristics of classification decisions.Results. According to the experimental data obtained from the electromyographic sensor, an input structured learning matrix for the alphabet with four recognition classes is formed. The decision rules constructed in the process of hierarchical informational and extreme machine learning make it possible to recognize cognitive teams in real time with a rather high total probability of making correct classifying decisions. The results of physical modeling proved that when using a hierarchical data structure in the form of a decursive tree, the functional efficiency of machine learning increases in comparison with the data structure in the form of a dichotomous binary tree. Conclusions. The results of physical modeling confirmed a sufficiently high functional efficiency of the proposed method of information-extreme machine learning for the control system of the prosthesis of the wrist with a non-invasive system for reading biosignals. The obtained scientific results open up a new direction in the creation of intellectual prostheses of the hand with a noninvasive system for reading biosignals based on machine learning and pattern recognition.