Identification of Durian Leaf Diseases Using the Residual Network Algorithm (RESNET-50)

Identifikasi Penyakit Daun Durian Menggunakan Penerapan Algoritma Residual Network (RESNET-50)

Abstract

Durian is one of Indonesia’s top horticultural commodities, but its productivity can decline due to leaf diseases that are difficult for farmers to identify visually. This study aims to develop an automated durian leaf disease classification system using a deep learning algorithm based on the ResNet-50 architecture. The dataset consists of 420 durian leaf images categorized into four classes: Algal Leaf Spot, Leaf Blight, Leaf Spot, and No Disease, sourced from the Roboflow platform. Preprocessing steps included annotation, augmentation, and resizing to 240x240 pixels. The model was trained using TensorFlow with pretrained ImageNet weights. Three data split scenarios (70:20:10, 75:15:10, and 80:10:10) were tested using both binary and multiclass classification approaches. Model performance was evaluated using a confusion matrix and metrics such as accuracy, precision, recall, and F1-score. The best binary classification achieved 99.8% accuracy and 99.9% F1-score, while the best multiclass model reached 99.6% accuracy and 96.9% macro F1-score.