Implementation of MobileNetV2-Based Deep Learning for Corn Leaf Disease Detection Using a Web-Based System

Abstract

This study investigates the hypothesis that a lightweight Convolutional Neural Network (CNN), specifically MobileNetV2, can achieve competitive classification accuracy for corn leaf disease detection while maintaining computational efficiency suitable for real-time web-based deployment. Corn is a critical agricultural commodity, and plant diseases significantly threaten productivity, causing economic losses and reduced yields. Traditional identification methods depend on manual observation, which is time-consuming, subjective, and error-prone at scale. While recent deep learning approaches such as Vision Transformers and hybrid architectures have demonstrated high accuracy, their computational demands limit practical deployment in resource-constrained agricultural environments. To address this gap, this research employs MobileNetV2 with transfer learning from ImageNet, leveraging depthwise separable convolutions for efficient feature extraction. The dataset comprises 3,852 corn leaf images categorized into four classes: Healthy, Common Rust, Gray Leaf Spot, and Northern Leaf Blight. Preprocessing includes image resizing to 224×224 pixels, normalization, and 80:20 train-validation splitting. The model was trained using TensorFlow and deployed via a Streamlit-based web interface for real-time inference. Experimental results demonstrate a training accuracy of 96.76% and a validation accuracy of 92.86%, with inference confidence reaching 99.80% during testing. The primary contribution of this work lies in the empirical demonstration that MobileNetV2, without architectural modification, can serve as a practical and accessible diagnostic tool when integrated into a web-based deployment pipeline, offering a favorable trade-off between accuracy and computational cost compared to heavier architectures.