IMPROVEMENT OF OBJECT DETECTION AND CLASSIFICATION ALGORITHMS IN COMPUTER VISION SYSTEMS

Abstract

This article analyzes the issues of improving object detection and classification algorithms in computer vision systems. As a result of the rapid development of modern artificial intelligence technologies, especially deep learning methods, the possibilities of accurately separating objects from image and video streams are expanding. The article considers the principles of operation of existing convolutional neural networks (CNN), region-based detection models, as well as their advantages and limitations. New approaches aimed at increasing accuracy, improving computing speed, and reducing resource consumption are also described. Data set expansion, image preprocessing, transfer learning, and the use of ensemble models are shown as important factors for improving efficiency. The practical significance of such improved algorithms in areas such as real-time surveillance systems, medical image analysis, industrial control, and autonomous transportation is highlighted. The research results show that the reliability of detection and classification processes is significantly increased by using optimized model architectures and adaptive training strategies.