Hybrid CNN and Machine Learning Framework for Detecting Curvularia Leaf Spot in Oil Palm Seedlings Using VGG16 and ResNet50

This study presents a hybrid framework for detecting black leaf spot disease in oil palm seedlings caused by Curvularia infection by combining deep learning with classical machine learning classifiers. A total of 5,141 images of healthy leaves and leaves exhibiting black spot symptoms were collected in September 2024 from an oil palm nursery in Sepakat Village, Masamba District, North Luwu Regency, South Sulawesi. All images were standardized through resizing, cleaning, and augmentation to increase visual variability and support more robust feature learning. Features extracted from the pre-trained VGG16 and ResNet50 networks were classified using Support Vector Machine, K-Nearest Neighbors, Naive Bayes, Decision Tree, and Random Forest algorithms. Three configurations, namely VGG16 with SVM, VGG16 with KNN, and ResNet50 with Random Forest, yielded the strongest performance across all evaluated metrics. While these results demonstrate the potential of combining convolutional feature extraction with conventional classifiers for leaf disease detection, they may also reflect limitations associated with data collected from a single location, which introduces risks of limited generalization. Although the models achieved inference speeds below 0.05 seconds per image, further validation using more diverse datasets and real-device testing on IoT or mobile platforms is required to assess their practical reliability.