Comparative Evaluation of Pre-Trained Deep Learning Models for Precision Diagnosis of Potato Leaf Diseases

Abstract

Diagnosing plant diseases is a tedious and time-consuming process that requires the expertise of professionals, whose success rate largely depends on their experience. Therefore, early and accurate diagnosis of plant diseases is important. Deep learning (DL) methods have been employed to tackle this challenge by developing automated detection systems capable of rapidly and accurately identifying plant diseases. Using the Potato Disease Leaf dataset and the fine-tuning concept, we compared five pre-trained models in this study: Resnet50, Xception,, Mobilenet, DenSeNet121, and EfficientNetB0. Among the pre-trained models, EfficientNetB0 performed the best, outperforming the others with equivalent accuracy values of 0.972, 0.971, 0.974, and 0.972 for precession, recall, and F1-Score.