Title: Deep Learning-Based Corn Disease Classification Using VGG16 Transfer Learning
Authors: Alaa Kh. I. AlDammagh, Samy S. Abu-Naser
Volume: 10
Issue: 2
Pages: 19-27
Publication Date: 2026/02/28
Abstract:
Corn (maize) is a critical global crop, yet its production is significantly threatened by various leaf diseases, leading to substantial economic losses and food security concerns. Traditional manual disease detection methods are time-consuming, subjective, and impractical for large-scale agricultural applications. This study presents a deep learning-based approach for automated classification of corn leaf diseases using transfer learning with the VGG16 convolutional neural network architecture. The dataset comprises 5,976 images across four categories: Blight (1,303 images), Common Rust (1,047 images), Gray Leaf Spot (1,643 images), and Healthy leaves (1,983 images), standardized to 256×256pixel resolution. The methodology employs a two-phase training strategy: initial training with a frozen VGG16 base followed by fine-tuning of the last eight layers using a reduced learning rate. Real-time data augmentation techniques including random rotations, width and height shifts, and horizontal and vertical flipping were applied to enhance model generalization and mitigate overfitting. The model achieved an overall test accuracy of 95.52%, correctly classifying 1,419 out of 1,494 test images. Per-class evaluation demonstrated exceptional performance for Healthy leaves (100% precision, 99.80% recall) and Common Rust (96.59% precision, 97.70% recall), with F1-scores of 0.9990 and 0.9714, respectively. The Blight and Gray Leaf Spot classes achieved F1-scores of 0.9053 and 0.9318, with minimal confusion attributed to visual similarities between lesion patterns. The minimal gap between training and validation loss curves confirmed robust generalization without significant overfitting. These findings demonstrate that transfer learning with VGG16 provides a reliable, computationally efficient solution for automated corn disease diagnosis, offering practical potential for deployment in resource-constrained agricultural settings through mobile or smartphone-based applications to support farmers in early disease intervention and crop management.