Klasifikasi Penyakit Batang Buah Naga Merah Menggunakan Arsitektur Eca-Resnet-50 Berbasis Website

Abstract

Red dragon fruit is one of the horticultural commodities with high economic value in Indonesia. One of the main challenges in dragon fruit cultivation is stem disease infection, which can reduce plant growth and crop productivity. Disease identification is still commonly performed through visual observation, making the process time-consuming and dependent on the observer's experience. This study aims to analyze the performance of the ECA-ResNet-50 model for red dragon fruit stem disease classification and implement it in a web-based system. The dataset consisted of six classes, namely Anthracnose, Brown Stem Spot, Gray Blight, Soft Rot, Stem Canker, and Healthy. The research data were obtained from a secondary dataset and a combined dataset consisting of secondary data and primary field documentation collected in Banyuwangi. The research stages included data collection, preprocessing, model training, evaluation using a confusion matrix, heatmap visualization using Grad-CAM, and system implementation using Flask. The experimental results showed that the best model was ECA-ResNet-50 using the Adam optimizer, batch size of 32, and the combined dataset, achieving an accuracy of 98.89%, precision of 98.96%, recall of 98.89%, and F1-score of 98.89%. The confusion matrix results indicated that 89 out of 90 test images were correctly classified. Grad-CAM visualization showed that the model focused on disease-affected regions of the dragon fruit stem during the classification process. The model was then deployed into a web-based application and demonstrated good performance in classifying images that had not been used during the training process. The results indicate that the addition of the ECA module improves classification performance and produces a system that can assist in the identification of red dragon fruit stem diseases.