Citra Drone RGB untuk Identifikasi Kerusakan Tanaman Padi (Oryza sativa L.) Akibat Organisme Pengganggu Tumbuhan (OPT)

Abstract

Tanaman padi (Oryza sativa L.) merupakan komoditas pangan utama Indonesia yang rentan terhadap serangan organisme pengganggu tumbuhan (OPT), salah satunya Hawar Daun Bakteri (BLB) yang disebabkan oleh bakteri Xanthomonas oryzae pv. oryzae. Pemantauan OPT secara konvensional memiliki keterbatasan dalam cakupan dan efisiensi waktu, sehingga diperlukan pendekatan penginderaan jauh berbasis drone RGB sebagai alternatif yang lebih cepat dan objektif. Penelitian dilaksanakan pada 31 Januari 2026 di Desa Hegarmanah, Kecamatan Bojongpicung, Kabupaten Cianjur, Jawa Barat. Citra drone RGB diperoleh menggunakan drone DJI Mavic 3 Enterprise (M3E) pada ketinggian ±15 m. Sebanyak 50 titik sampel validasi ditentukan secara sistematik diagonal dan random. Tingkat keparahan BLB diklasifikasikan menjadi empat kelas, yaitu sehat, ringan, sedang, dan berat. Nilai spektral RGB dan indeks vegetasi VARI, GLI, dan NDGI diekstrak menggunakan Zonal Statistics dan raster calculator di QGIS, kemudian diklasifikasikan menggunakan algoritma Random Forest (RF) secara terpisah maupun komposit. Nilai median RGB meningkat konsisten seiring meningkatnya keparahan BLB. Klasifikasi berbasis NDGI menghasilkan akurasi tertinggi (Overal Accuracy 94%; Kappa Accuracy 0,90), diikuti GLI (92%; 0,87), RGB (90%; 0,833), dan VARI (88%; 0,80), sedangkan klasifikasi komposit seluruh variabel spektral dan indeks vegetasi menghasilkan akurasi terendah (84%; 0,73) akibat Redundansi informasi antarvariabel yang saling berkorelasi. Hasil ini menunjukkan bahwa NDGI merupakan variabel input paling efektif untuk klasifikasi tingkat kerusakan tanaman padi akibat BLB berbasis citra drone RGB.

Rice (Oryza sativa L.) is a major food commodity in Indonesia and is susceptible to attack by plant pests and diseases, one of which is bacterial leaf blight (BLB) caused by the bacterium Xanthomonas oryzae pv. oryzae. Conventional monitoring of plant pests and diseases has limitations in terms of coverage and time efficiency; therefore, a remote sensing approach using RGB drone imagery is needed as a faster and more objective alternative. This study was conducted on January 31, 2026, in Hegarmanah Village, Bojongpicung District, Cianjur Regency, West Java. RGB drone imagery was acquiRed using a DJI Mavic 3 Enterprise (M3E) at an altitude of approximately 15 m. A total of 50 validation sample points were determined using systematic diagonal and random sampling methods. BLB severity was classified into four classes: healthy, light, moderate, and severe. RGB spectral values and the vegetation indices VARI, GLI, and NDGI were extracted using Zonal Statistics and Raster Calculator in QGIS and subsequently classified using the Random Forest (RF) algorithm, both individually and in combination. The median RGB values increased consistently with increasing BLB severity. Classification based on NDGI achieved the highest Accuracy, with an Overall Accuracy of 94% and a Kappa coefficient of 0.90, followed by GLI (92%; 0.87), RGB (90%; 0.83), and VARI (88%; 0.80). In contrast, the classification using a composite of all spectral variables and vegetation indices resulted in the lowest Accuracy (84%; 0.73), which was attributed to Redundant information among correlated variables. These results indicate that NDGI is the most effective input variable for classifying the severity of BLB damage in rice plants using RGB drone imagery