A Hybrid Transfer Learning Framework for Automated Potato Leaf Disease Classification Using Lightweight Convolutional Neural Networks
Sangeeta Jana Mukhopadhyay, Arighna Basak and Angshuman Majumdar
Potato (Solanum tuberosum L.) is one of the world’s most popular food crops, and the early detection of leaf diseases is essential for minimizing yield losses and ensuring sustainable agricultural production. Traditional disease diagnosis techniques are often labour-intensive, time-consuming, and prone to subjective errors. This paper focuses on a hybrid transfer learning framework based on MobileNetV3-Small for the automated classification of potato leaf diseases. The proposed model integrates image preprocessing, transfer learning, data augmentation, and Gradient-weighted Class Activation Mapping (Grad-CAM) within a unified lightweight architecture to improve classification accuracy with enhanced model interpretability. The framework was evaluated using Receiver Operating Characteristic (ROC) curves, accuracy, precision, recall, F1-score, and confusion matrix analysis. Experimental results demonstrated an overall classification accuracy of 95.8%, exhibiting reliable discrimination between healthy and diseased potato leaves. Furthermore, Grad-CAM visualizations effectively highlighted disease-affected regions, thereby improving the transparency and interpretability of the classification methods. The combination of lightweight architecture, explainable artificial intelligence, and robust predictive performance makes this model well-suited for deployment on mobile and edge computing platforms with an efficient decision-support tool for precision agriculture and intelligent crop monitoring applications.
Keywords: Potato Leaf Disease Detection, Deep Learning, Convolutional Neural Network, Explainable Artificial Intelligence, Grad-CAM, Precision Agriculture, Image Processing