A Comparative Analysis of Convolutional Neural Networkbased Transfer Learning Models for Plant Disease Detection

Abstract

Agriculture holds a significant role in the economy of many countries by providing food security, employment opportunities, and contributing to the overall economic growth of the nation. However, the spread of plant diseases can pose a significant threat to crop production and result in reduced yields, food scarcity, and ecological imbalances. The current manual process of diagnosing plant diseases is not only time-consuming but also prone to errors, relying heavily on the expertise of pathologists. To tackle this challenge, this paper presents a user-friendly and accessible machine learning-based plant disease detection system that uses Convolutional Neural Network (CNN) based transfer-learning models to detect and categorize plants as either healthy or diseased and provides information on the precautions for the identified disease. By comparing the performance of three cutting-edge CNN architectures, VGG16, ResNet50, and EfficientNetV2S, on publicly available plant disease dataset, this paper aims to determine a model for detecting and classifying various plant diseases. We found that ResNet50 achieves the highest accuracy of 96.5%. This proposed system can enable timely and accurate disease detection, helping farmers take preventive measures and reducing the risk of crop loss.