An Efficient Transfer Learning Approach for Identification and Severity Prediction of Plant Diseases

Plant diseases significantly impact agricultural yield, raising serious concerns for sustainable farming. Early and accurate disease detection is essential to prevent crop losses and ensure food security. This study introduces a deep learning–based model, PDISA, designed to detect and classify tomato leaf diseases using EfficientNet and ResNet architectures within a CNN framework. Among the two, EfficientNet demonstrated superior performance, achieving 92% precision and 89% accuracy. The model leverages transfer learning, utilizing the PlantVillage dataset, where the feature extraction module is frozen and the classification layers are finetuned to improve accuracy. This paper also proposes a severity assessment module based on segmentation techniques to sort disease advancement into four specific tiers (0-3, where 0 is healthy leaves and 3 is heavy infection), along with ARIMA modeling for rotting time predictions, all of which are being made accessible on a mobile platform with chatbot interface for instant disease examination and treatment suggestion.