Potato Leaf Disease Classification Using Lightweight CNN-SVM and Real-World Web-Based Tools

Potatoes play an important role as a major agricultural product since the economic system of Bangladesh strongly relies on farming. The presence of plant diseases particularly late blight poses substantial threats to economic stability together with food security and harvest yields. Lightweight Convolutional Neural Network (CNN) and Support Vector Machine (SVM) represent a new approach to automate the detection of diseases appearing on potato leaves. The validation accuracy obtained by this model reaches an average of 98.53%. The model sustains high performance by employing 0.45 million parameters which leads to minimized storage needs equivalent to 1.75 MB and reduced computational expenses. SHAP (SHapley Additive exPlanations) together with LIME (Local Interpretable Model-agnostic Explanations) serve as explainable AI techniques used to enhance model prediction transparency. The proposed model excels beyond VGG-16 and VGG-19 transfer learning standards in all evaluation metrics including precision, recall and F1-score even though its parameter count stands at 32.8× and 44.6× lower. Mobile deployment possibilities of the simplified modelarchitecture allow farmers to receive prompt disease diagnoses along with correct decision-making resources. A web-based platform enables users to analyze potato leaf diseases by uploading images in real-time. The study creates substantial progress in agricultural productivity alongside ensuring food security for Bangladesh.