Sistem deteksi hama dan penyakit daun kelapa dengan metode CNN menggunakan arsitektur MobileNetV2

Abstract

Coconut plants are a vital source of income for communities in East Melonguane District, yet they are vulnerable to leaf pests and diseases that disrupt photosynthesis and reduce crop yields. Farmers often detect symptoms too late, resulting in delayed treatment and increased damage. Although previous studies using Convolutional Neural Networks (CNN) have demonstrated promising accuracy, only a few have implemented web-based systems specifically for coconut leaves in remote areas, and many lack detailed hyperparameter reporting for reproducibility. Therefore, this study aims to develop a web-based application for detecting coconut leaf pests and diseases using a CNN with the MobileNetV2 architecture. A dataset consisting of 1,000 images across four categories—Sitora nitens, dry leaves, wilted leaves, and yellowing leaves—was used. The model was trained using transfer learning with 25 epochs, a batch size of 16, a learning rate of 0.0001, the Adam optimizer, and categorical cross-entropy loss. The experimental results show that the proposed model achieved an overall accuracy of 96% (193 out of 200 test samples correctly classified), with a precision of 96%, recall of 96%, and F1-score of 96%, indicating its effectiveness for fast and accurate classification in practical applications.