DESIGN AND IMPLEMENTATION OF A PLANT DISEASE DETECTION SYSTEM USING MACHINE LEARNING

ABSTRACT This study presents the design and implementation of a plant disease detection system using machine learning, exploring two distinct approaches to improve accessibility and accuracy in disease diagnosis. The first approach employs Google’s Gemini, a generative AI model, integrated with a mobile application built in React Native. This app allows users to take photos of plants, which Gemini then analyzes to detect potential diseases. This method was chosen due to the limited availability of datasets specific to indigenous crops, and it benefits from Gemini's extensive knowledge base, which covers a wide range of plant diseases. However, Gemini's 70% accuracy on well-known plant diseases is limited by the fact that generative models are trained on vast, diverse datasets that may include unverified information. The second approach uses a Convolutional Neural Network (CNN) combined with OpenCV to analyze images for plant defects. While this method achieves higher accuracy, it is constrained by the availability of high-quality, targeted datasets, which limits its applicability to a narrower range of diseases. This dual approach highlights both the versatility of generative AI for broad, generalizable diagnosis and the enhanced precision of CNN-based models when sufficient data is available.