Corn Leaf Disease Identification via Transfer Learning: A Comprehensive Web-Based Solution

Abstract

Effective crop disease prevention is essential to ensure global food security and early disease detection is a vital part of this protection. Traditional techniques of identifying disease are lengthy process, costly, sometimes require specialized knowledge, and nevertheless may produce erroneous outcomes. Artificial intelligence offers the best answer in this situation. Deep learning has become essential for analyzing images and classification. This study proposes a website that uses deep learning for classifying three major diseases of maize leaves: blight, common rust, and grey leaf spot as well as for identifying healthy leaves. Additionally, it conducts a comparative analysis of various state-of-the-art models using the same dataset to determine the most suitable approach for website development considering metrics such as accuracy, precision, recall, F1 score, training time, and model size. All the used models (MobileNetV2, AlexNet, ResNet18, VGG16, VGG19, and SqueezeNet) have been optimized for faster operation and lower storage consumption. The models were trained using the “Corn or Maize Leaf Disease Dataset” on Kaggle, which included 2930 images of maize leaves. After that, the models were tested using a separate set of 422 images, categorized into four classes: three representing diseases (blight, common rust, and grey leaf spot) and the fourth representing healthy leaves. Out of all the models, ResNet18 has the highest accuracy (96.45%). ResNet18 has several evaluation matrices that make it ideal for this investigation, including quick training and a small model size. As ResNet18 provides the best result, the website can accurately classify disease class and display the probability of identification for uploaded corn leaf images using this model. The model's performance is found satisfactory for its real-world application in automatically detecting maize leaf diseases.