Порівняльний аналіз моделей комп’ютерного зору для розпізнавання рослин

Abstract

The paper presents a comparative analysis of computer vision models for automatic plant recognition based on images of leaves, flowers, and stems. The research is motivated by the growing importance of digital agriculture and environmental monitoring systems that require accurate and efficient classification methods. The study considers convolutional neural networks (CNN) as the fundamental approach for feature extraction and image classification. Mathematical principles of convolution operations and classification accuracy metrics are described to evaluate model performance. Three popular architectures are analyzed: classical CNN, ResNet, and EfficientNet. Each model is compared in terms of accuracy, computational complexity, and suitability for practical applications. The results demonstrate that lightweight architectures are preferable for mobile and embedded systems, while deeper networks provide higher accuracy for scientific and cloud-based solutions.