MASYKUR, Fauzan and Adi, Kusworo and Nurhayati, Oky Dwi (2024) PENGEMBANGAN ALGORITMA SEGMENTASI KONTUR DAN YOU ONLY LOOK ONCE (YOLO) UNTUK MENGHITUNG LUAS WILAYAH TANAMAN PADI TERINFEKSI DENGAN MENYISIPKAN CITRA REFERENSI ARUCO MARKER. Doctoral thesis, UNIVERSITAS DIPONEGORO.

	Text cover DISERTASI.pdf Download (2MB)
	Text BAB I.pdf Download (215kB)
	Text BAB II.pdf Download (5MB)
	Text BAB III.pdf Restricted to Repository staff only Download (19MB)
	Text BAB IV.pdf Restricted to Repository staff only Download (26MB)
	Text BAB V.pdf Restricted to Repository staff only Download (231kB)
	Text DAFTAR PUSTAKA.pdf Download (193kB)
	Text LAMPIRAN.pdf Restricted to Repository staff only Download (688kB)

Abstract

Adanya teknologi artificial intelligence (AI) merubah paradigma dan merevolusi dunia pertanian dalam melindungi hasil panen dari beberapa factor seperti perubahan iklim, pertumbuhan penduduk hingga serangan hama dan penyakit. Sepertinya halnya pada tanaman padi yang tidak luput dari serangan hama dan penyakit yang bisa mempengaruhi hasil panen. Adanya penyakit pada tanaman padi tersebut diperlukan adanya deteksi penyakit tanaman padi secara cepat, akurat dan tepat. Sebelum adanya artificial intelligence deteksi penyakit tanaman padi dilakukan dengan pemantauan secara langsung oleh petani yang berpengalaman dan membutuhkan waktu serta hasil pengamatan sangat mungkin berdasarkan persepsi subjektif masing-masing petani. Kecepatan deteksi penyakit tanaman padi secara dini bisa dilakukan dengan memanfaatkan citra udara dari kamera drone sebagai inputan kemudian dideteksi objek keberadaan penyakit dengan mengembangkan algoritma deteksi objek You Only Look Once (yolo) dan segmentasi kontur.
Menggunakan drone atau Unmanned Aerial Vehicle (UAV) sebagai alat akuisisi citra yang terbang di atas lahan persawahan padi untuk meng-capture lahan sebagai dataset deteksi objek. Dataset yang terkumpul dilabeli menggunakan software yolo-mark dan labelImg dengan 2 kelas yakni kelas infected dan healthy. Wilayah persawahan tanaman padi yang terinfeksi penyakit dihitung luasnya dengan menyisipkan citra acuan atau referensi. Citra referensi berupa ArUco Marker ini digunakan sebagai acuan dalam penentuan awal koordinat sebagai langkah awal menghitung panjang piksel. Setelah piksel diketahui dikonversi ke satuan panjang (meter) sehingga diketahui luas wilayah persawahan terinfeksi penyakit.
Pelatihan model deteksi objek yolo dilakukan setelah pelabelan yang menghasilkan nilai akurasi terbaik senilai 77,3 % pada jarak akuisisi 20 meter dan nilai terendah senilai 46,8 % pada jarak akuisisi 2 meter. Hasil prediksi penyakit pada tanaman padi ini di validasi oleh petugas pertanian yang menghasilkan nilai 97,02 % hasil prediksi sama dengan kondisi nyata dan 2,80 % prediksi tidak sama dengan kondisi nyata. Sedangkan perhitungan luas wilayah tanaman yang terdeteksi dihitung dengan menyisipkan citra referensi ArUco Marker menghasilkan perbedaan luas antara prediksi menggunakan yolo dan kondisi nyata. Selisih antara luas prediksi yolo dan luas nyata memiliki nilai terkecil senilai 0,3m2 pada citra hasil akuisisi jarak 5 meter dan nilai terbesar pada jarak 20 meter senilai 35,36 m2. Perbandingan perhitungan luas wilayah terinfeksi antara perhitungan berdasarkan hasil deteksi objek yolo dengan segmentasi kontur menghasilkan nilai yang lebih baik pada segmentasi kontur. Nilai selisih terkecil antara perhitungan segmentasi dengan luas nyata senilai 0,22 m2 yang terjadi pada file citra 0824_10m. Sedangkan pada perhitungan luas berdasarkan hasil deteksi yolo nilai selisih terkecil senilai 3,77 m2 yang terjadi pada file citra 0754_10m.
Kata kunci : Deteksi objek, yolo, segmentasi kontur, ArUco Marker, tanaman padi

The existence of artificial intelligence (AI) technology is changing the paradigm and revolutionizing the world of agriculture by protecting harvests from several factors, such as climate change, population growth, and pest and disease attacks. It seems that rice plants are not immune from pests and diseases that can affect crop yields. The presence of disease in rice plants requires rapid, accurate, and precise detection of rice plant diseases. Before the existence of artificial intelligence, detection of rice plant diseases was carried out by direct monitoring by experienced farmers and took time, and the results of observations were very likely based on the subjective perception of each farmer. The speed of early detection of rice plant diseases can be achieved by using aerial images from drone cameras as input and then detecting the presence of disease objects by developing the You Only Look Once (yolo) object detection algorithm and contour segmentation.
Using a drone or Unmanned Aerial Vehicle (UAV) as an image acquisition tool that flies over rice fields to capture the land as an object detection dataset. The collected dataset is labeled using yolo-mark and labelImg with 2 classes, namely infected and healthy classes. The area of rice fields infected with disease is calculated by inserting a reference image. The reference image in the form of the ArUco Marker is used as a reference in determining the initial coordinates as the first step in calculating the pixel length. Once the pixels are known, they are converted to units of length (meters), so that the area of the rice fields infected with the disease is known.
The Yolo detection model was trained, which produced the best accuracy value of 77.3% at an acquisition distance of 20 meters and the lowest value of 46.8% at an acquisition distance of 2 meters. Then the results of disease predictions in rice plants were validated by agricultural officers, who produced a value of 97.02% of the predicted results were the same as real conditions, and 2.80% of the predictions were not the same as real conditions. Meanwhile, the calculation of the area of the detected plant area calculated by inserting the ArUco Marker reference image produces a wide difference between predictions using Yolo and real conditions. The difference between the Yolo predicted area and the actual area has the smallest value of 0.3 m2 in the image acquired at a distance of 5 meters, and the largest value at a distance of 20 meters is 35.36 m2. A comparison of the calculation of the area of the infected area between calculations based on the results of Yolo object detection and contour segmentation produces better values for contour segmentation. The smallest difference between the segmentation calculation and the real area is 0.22 m2, which occurs in image file 0824_10m. Meanwhile, in the area calculation based on the Yolo detection results, the smallest difference value was 3.77 m2, which occurred in image file 0754_10m.
Keywords: object detection, yolo, contour segmentation, ArUco Marker, rice plants

Actions (login required)

View Item