Название: Artificial Intelligence in Models, Methods and Applications
Автор: Olga Dolinina, Igor Bessmertny, Alexander Brovko, Vladik Kreinovich
Издательство: Springer
Год: 2023
Страниц: 694
Язык: английский
Формат: pdf (true)
Размер: 21.3 MB

This book presents new perspective research results: models, methods, algorithms and applications in the field of Artificial Intelligence (AI). Particular emphasis is given to the medical applications - medical images recognition, development of the expert systems which could be interesting for the AI researchers as well for the physicians looking for the new ideas in medicine. The central audience of the book are researchers, industrial practitioners, students specialized in the Artificial Intelligence.

The Chapter "Development of a Program Code Review System Using Machine Learning Methods" presents a description of the developed approach and service for analyzing source code in Python. The service reduces the time for code review due to partial automation. The FastText algorithm is used to obtain vector representations of source code texts. A pre-trained neural network language model based on the transformer architecture was used to derive a possible natural language function assignment. A classifier based on the gradient boosting algorithm was used to detect duplicate PR. The developed service checks the changeset and publishes error and duplicate reports in changeset comment format after the changeset is published to a remote Git repository. The conducted testing did not reveal any errors that affect the operation. All the main functions of the system are performed correctly.

The Chapter "Production Control Based on a Quality Guarantor Computer Vision System" presents a solution for an intelligent quality control system capable of identification of product failures and manufacturing defects using a computer vision system. The proposed solution expands the capabilities of automated quality management systems in modern production by supporting operator actions using an artificial neural network. The resulting solution is called “intelligent quality guarantor”, being proposed and described in this paper it implements a new concept of production process multiple parameters monitoring and recognition using an intelligent computer vision system. Special software was developed capable of dynamic tracking of quality control process to decompose and classify the tasks being performed by patterns of deviance. The computer vision application being developed collects real-time data from cameras and analyzes data streams using machine learning algorithms. Based on predetermined quality standards, the system is capable of automatic sorting, as well as detecting visual defects corresponding to mechanical damage or manufacturing defects.

The Chapter "Deep Learning Approach to Recognition of Car Driver's Closed Eyes for Safety Reasons": Today it is difficult to overestimate the role of artificial intelligence systems in various sectors of the economy. One of the rapidly developing areas in the period of digitalization is the sphere of transport security. The development of unmanned vehicles is promising, but such systems raise many questions about the perpetrators in the event of various accidents. In this regard, at present, the vast majority of cars are driven by a person, which leads to corresponding problems. One of the key problems is the high proportion of accidents associated with falling asleep at the wheel. This article is aimed at solving the problem of analyzing the driver's condition using a video camera installed in the car. The system uses approaches from both traditional machine learning methods and deep learning methods based on convolutional neural networks. To detect eyes with a completeness close to 100%, in the absence of interfering objects, Haar cascades are proposed. For this, simple geometric calculations are performed based on the detected face, which significantly speeds up the process. In this case, a parallel system of two convolutional networks is used to recognize closed and open eyes. The completeness of recognition of closed eyes in one frame in the developed system is almost 90%.