The dissertation is devoted to increasing the accuracy of the execution of cognitive commands by prosthetic limbs by creating an information technology of machine learning of the hand prosthesis control system for recognition of EMG biosignals within framework of a functional approach to modeling cognitive processes of natural intelligence in forming and making classification decisions. According to the results of analysis of the current state and trends in development of intelligent limb prostheses, it is shown that prostheses with an invasive system for recording EMG biosignals are characterized by greater accuracy in the execution of cognitive commands in comparison with a non-invasive system. In contrast invasive biosignal recording system requires surgical intervention, creates uncomfortable conditions for a person with a disability, and does not ensure the restoration of lost hand functions due to the imperfection of existing information technologies for recognizing EMG signals. Prostheses with a non-invasive recording system are significantly cheaper and more convenient to use compared to hand prostheses with an invasive biosignal recording system. Due to high noise of biosignals, the multidimensionality of recognition features space and the significant intersection in the features space of recognition classes that characterize main movements of the prosthesis, the development of intelligent prostheses requires overcoming scientific and methodological complications, which are influenced by the following factors: arbitrary initial conditions for the formation of EMG biosignals caused by damage to cognitive-nerve path of biosignal transmission; recognition classes intersection, which creates an a priori unclear division of the feature space; multidimensionality of the feature dictionary and the alphabet of recognition classes; the influence of uncontrollable factors related to density of attachment of EMG sensors, their position, changes in power parameters, the emotional and mental state of a person with a disability, etc., which causes noise, artifacts and distortion of EMG biosignals of the corresponding cognitive commands. It is for these reasons that the machine learning algorithms of the hand prosthesis control system within known methods of Data Mining technology, including neuro-like structures, do not ensure high accuracy of cognitive command execution. Therefore, the dissertation, which was carried out in the scientific laboratory of the Department of Computer Sciences of Sumy State University, is relevant, as it is aimed at solving an important scientific and practical task of increasing the functional efficiency of hand prostheses with a non-invasive biosignal recording system and bringing them closer to the functional capabilities of invasive while maintaining a relatively low cost is relevant. In the dissertation, the research was carried out within the framework of the information-extreme intelligent technology of data analysis created at Sumy State University, which is based on maximizing the information capacity of control system in the process of machine learning. The idea of the developed methods of information-extreme machine learning of the hand prosthesis control system for recognition of EMG biosignals, as well as in artificial neural networks approach, is to adapt the input mathematical description to the maximum full probability of making correct diagnostic decisions in the process of machine learning. Despite this the main advantage of information-extreme machine learning methods over neuro-like structures is that they are developed within the framework of a functional approach to modeling cognitive processes inherent in humans when forming and making classification decisions. This approach, unlike artificial neural networks, allows the methods of IEIT machine learning to provide flexibility in retraining the system through the expansion of the alphabet of recognition classes. At the same time, decisive rules constructed within the framework of geometric approach are practically invariant to multidimensionality of the dictionary of recognition features. In addition, the formation of the training matrix requires an far fewer samples, which is an important advantage over neuro-like structures. In the dissertation work, an important scientific and practical task of developing the information intelligent technology of machine learning of the hand prosthesis control system with a non-invasive system for recording biosignals within the framework of a functional approach to modeling cognitive processes is solved. On the basis of proposed models, methods and algorithms, the means of information technology of machine learning of the hand prosthesis control system have been implemented, which include modules for the formation of input mathematical description, machine learning, construction of decisive rules and functioning systems in the modes of functional testing and examination.
