Image Classification Using MobileNet Based on CNN Architecture for Grape Leaf Disease Detection

Abstract

Grape cultivation, while economically important, is often challenged by various leaf diseases that can significantly impact yield and quality, underscoring the need for rapid and accurate detection methods. Traditional diagnostic approaches can be time-consuming and require expert knowledge, whereas advanced image classification techniques offer a promising avenue for automated disease identification. This research aimed to develop and rigorously evaluate a Convolutional Neural Network (CNN) model, specifically leveraging the MobileNetV2 architecture, for the precise classification of four common grape leaf diseases: healthy, Black Rot, Esca (also known as Black Measles), and Leaf Blight. The methodology encompassed dataset acquisition and pre-processing, data augmentation to increase training data diversity, and applying transfer learning using pre-trained MobileNetV2 weights, followed by a fine-tuning stage to adapt the model to the specific task. A comprehensive evaluation on 1,805 previously unseen test images demonstrated the model's exceptional performance, achieving an overall accuracy of 99.89%. Ultimately, the proposed approach significantly outperforms previous methods, demonstrating the feasibility of applying lightweight CNN architectures to real-world detection scenarios. The main contribution of this research is showing that high computational efficiency can be achieved without sacrificing accuracy, paving the way for implementation in digital detection systems with limited resources, particularly for mobile devices or edge systems.