Automated identification of plant diseases from digital images using deep learning methods

Abstract

The article describes the development of an automated system for identifying plant diseases using machine learning techniques. The research is aimed at automating the recognition of plant diseases based on visible external symptoms. Its primary objective is to create an intelligent software solution capable of quickly and accurately evaluating plant health from photographic images.The study covers the key stages of designing and implementing an image-based plant disease recognition system. Existing applications for assessing plant condition were reviewed to determine their main advantages and limitations. Based on this analysis, the functional requirements and quality criteria for the proposed system were defined. Special emphasis was placed on convolutional neural networks, image processing techniques, feature extraction approaches, and the characteristics of various neural network architectures. During the experimental stage, several neural network models were trained using the PlantVillage and PlantWild datasets, and their classification performance was evaluated and compared. The model that achieved the best testing results was selected for integration into the proposed system. Its generalization capability was additionally assessed using the independent PlantDoc dataset. The resulting plant disease identification system was implemented as a mobile application that allows users to analyze a photograph of a plant and obtain information about a probable disease. The developed solution may serve as a practical tool for rapid plant health assessment and early detection of disease symptoms. Its use can support agricultural specialists in responding to potential problems more quickly and in taking timely measures to protect plants and reduce possible crop losses.