Deep Learning-Based Classification of Date Palm Leaf Health: A Comparative Study of Convolutional Neural Network Architectures

Abstract

Diagnosis of plant diseases, particularly in crops such as Date Palm, is crucial for preserving agricultural productivity and economic returns. This study addresses the challenge of accurately determining the health status of Date Palm leaves using deep learning models to classify them as healthy or diseased, focusing specifically on two prevalent conditions: brown spot disease and white scale infection. The dataset, sourced from Kaggle, comprises 2,631 images—470 of brown spot disease, 1,203 of healthy leaves, and 958 of white scale disease—categorized into three distinct classes. Three CNN-based architectures, namely GoogLeNet, AlexNet, and VGG19Net, were trained and evaluated to recognize the characteristic features of each class. Model performance was assessed using accuracy, sensitivity, specificity, F-measure, Positive Predictive Value, and Negative Predictive Value, with GoogLeNet achieving the highest accuracy (95.14%) and best specificity for distinguishing diseased from healthy leaves. By demonstrating the effective application of established CNN models in this domain, this work contributes to precision agriculture by offering an automated, high-precision approach for early disease diagnosis in Date Palms.