Development of Models for Searching Geon’s and Using Them in the Tasks of Identifying Objects and Surrounding Space from Photo Images

The work is devoted to the development of a computer vision algorithm for classifying types of image scenes according to a set of primitive objects on them. The implementation of a software system for testing the algorithm is described. The "Introduction" provides a justification for the relevance of studying an image classification algorithm based on the primitive components of objects on them. The object and subject of the study are defined, the purpose and objectives are set. The first chapter provides an analysis of the subject area of computer vision in robotics, an analysis of existing methods of object detection and image classification. Based on the analysis, the functional and user requirements for the system are revealed. The second chapter describes the process of designing an ensemble of detection and classification models to implement the algorithm. The process of forming a dataset for training models based on real and synthetic images is presented. The results of training detection models, the process of developing and testing classification models based on the GRU attention mechanism using the PyTorch library are presented. Neural network models are combined into an image classification algorithm. The third chapter presents the process of designing and developing a Python system for testing an algorithm. Based on the requirements identified at the analysis stage, use cases were identified and sequence diagrams in UML notation were formed. The fourth chapter contains the results of testing the algorithm on real-world photographs. The process of testing the software system using HTTP requests with the Postman tool is described. The work contains 87 pages of the main text, 50 pages of appendices, 53 figures and 12 tables. The bibliographic list includes 19 titles. "Development of Models for Searching Geon’s and Using Them in the Tasks of Identifying Objects and Surrounding Space from Photo Images", Belova A.V.