Kyslyi R. Recognition of types of human activity with the help of artificial intelligence

The dissertation is devoted to the research of methods of recognition of human activity and systems of data acquisition from body sensors in connection with the development of BSN technologies. Such technologies allow to constantly monitor the physiological parameters of a person, which in turn allows to detect exacerbations of chronic diseases, sudden crises, as well as to carry out long-term monitoring of human health. The methods of data collection and pre - processing from sensors are substantiated in the work. To pre-process such data, an algorithm has been developed that improves the filtering of noise arising during data collection. The algorithm involves converting one-dimensional signals (1d) of the accelerometer into two-dimensional (2d) graphics, using the Hamming window, stacking the signals together sequentially in rows (stacking), which allows each sequence of signals to correlate with other sequences. The optimality of such an approach is proved by comparing with others and measuring the accuracy of classification using models with the same architectures. The architecture of the neural network model is proposed for effective classification of human activity. For the first time, an approach was developed in the problem of determining human activity for the use of models trained on other data sets (so-called transfer training). Simulation of such training for the problem of classification of types of breath is carried out. A device with sensors capable of recording the movements of the chest has been built and a model of machine learning has been developed, which is able to classify human respiration on the basis of data from the device. Also in this work the discussed ability to use transfer training to the problem of human activity recognition. Created a simulation model for classification of breathing types. To test the hypotheses, was developed a device with sensors capable of recording the movements of the chest. Also, machine learning model for human respiration classification was trained, using data from the device.