Komparasi Evaluasi Status Kesehatan Tanaman Kelapa Sawit Berdasarkan VARI dan Spektrum RGB Menggunakan Model ANN

Abstract

Produktivitas kelapa sawit di Indonesia sangat dipengaruhi oleh kesehatan tanamannya. Deteksi dini kondisi kesehatan tanaman sangat krusial untuk mencegah kerugian ekonomi akibat penurunan hasil panen. Metode konvensional yang kurang efektif mendorong adopsi teknologi otomatisasi, seperti penginderaan jauh menggunakan drone. Meskipun drone dengan sensor multispektral sering digunakan, kamera RGB yang lebih ekonomis dan mudah dioperasikan memiliki potensi untuk deteksi dini kesehatan tanaman. Penelitian ini bertujuan mengevaluasi dan membandingkan akurasi model Artificial Neural Network (ANN) dalam mendeteksi status kesehatan tanaman kelapa sawit menggunakan dua jenis input, yaitu indeks VARI dan spektrum RGB. Penelitian ini menggunakan 3.058 sampel tanaman kelapa sawit dari data citra RGB dan multispektral dari perkebunan kelapa sawit di Kebun Percobaan Jonggol IPB University. Indeks NDVI digunakan sebagai data ground truth untuk klasifikasi empat kategori kesehatan, ‘sehat’, ‘cukup sehat’, ‘kurang sehat’, dan ‘kritis’. Model ANN dilatih menggunakan 500 sampel dan diuji pada 3.058 tanaman kelapa sawit. Hasil menunjukkan kedua model memiliki keunggulan masing-masing tergantung pada tujuan pemantauan. Model ANN berbasis VARI dengan akurasi tertinggi 99,67% lebih unggul untuk pemetaan makro karena mampu merepresentasikan distribusi kesehatan tanaman yang konsisten. Model RGB dengan akurasi 99,51% lebih sesuai untuk deteksi dini dan pemantauan spasial kondisi tanaman karena lebih sensitif pada gejala awal stres.

Oil palm productivity in Indonesia is significantly influenced by plant health. Early detection of plant health is crucial to prevent economic losses resulting from yield reduction. Conventional methods, which are less effective, have spurred the adoption of automation technologies, such as remote sensing using drones. Although drones equipped with multispectral sensors are commonly employed, more economical and easily operable RGB cameras hold potential for early plant health detection. This research aimed to evaluate and compare the accuracy of Artificial Neural Network (ANN) models in detecting oil palm health status using two types of input, the VARI index and the RGB spectrum. The study utilized 3,058 oil palm plant samples derived from RGB and multispectral imagery data obtained from an oil palm plantation at the Jonggol Experimental Farm of IPB University. The NDVI index was used as ground truth data for classifying four health categories, healthy, moderately healthy, less healthy, and critical. The ANN models were trained using 500 samples and tested on 3,058 oil palm plants. The results showed that both models had their respective advantages depending on the monitoring objectives. The VARI model, which achieved the highest accuracy of 99.67%, was more suitable for macro-level mapping due to its ability to consistently represent plant health distribution. The RGB model, which achieved an accuracy of 99.51%, was better suited for early detection and spatial monitoring of plant conditions because of its higher sensitivity to early stress symptoms.