Deteksi Dini Kebakaran Hutan Berbasis Citra Streaming Video Menggunakan Unmanned Aerial Vehicle Dengan Image Processing Dan Object Detection

Ardiansyah, Axcel Nirmaulana (2025) Deteksi Dini Kebakaran Hutan Berbasis Citra Streaming Video Menggunakan Unmanned Aerial Vehicle Dengan Image Processing Dan Object Detection. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5009211020-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (5MB) | Request a copy

Abstract

Kebakaran hutan berdampak luas terhadap ekosistem, kesehatan, dan ekonomi. Penelitian ini mengembangkan sistem deteksi berbasis Unmanned Aerial Vehicle (UAV) menggunakan pemrosesan citra object detection dengan YOLOv8s untuk meningkatkan efektivitas deteksi kebakaran. Sistem ini menggunakan kamera NOIR dengan pendekatan RGB yang sensitif terhadap sepktrum NIR pada kanal merah (R) untuk mengangkap emisi panas dari objek api. Citra dari UAV dikirim secara real-time menggunakan internet ke Ground Control Station (GCS) untuk diproses. Hasil penelitian menunjukkan sistem ini mampu mendeteksi kebakaran seluas 0,6 x 0,6 meter persegi hingga ketinggian terbang 90 meter pada semua variasi lokasi titik hotspot, meskipun bounding box YOLOv8s cenderung lebih besar dari ukuran api sebenarnya akibat faktor lidah api dan limitasi resolusi video. Pengukuran luas api menggunakan metode Ground Sample Distance (GSD) menunjukkan rentang eror 0,13 - 1,71 meter pada ketinggian 60 - 90 meter pada semua posisi deteksi, yang masih dapat diterima dalam konteks pengukuran objek dari udara berdasarkan standar ASPRS 2024 untuk RMSEH 0,75 - 1,5 meter. Geotagging lokasi api memiliki rentang eror 2,35 - 7,85 meter (60 meter) dan 2,4 - 8 meter (90 meter) akibat keterbatasan GNSS UAV, crosswind, serta offset antara kamera dan GNSS UAV tetapi sesuai standar dari GPS NAVSTAR. Kesimpulan yang didapat adalah metode RGB dengan sensitivitas NIR pada kanal merah menggunakan kamera NOIR dapat mendeteksi kebakaran kecil secara efektif, serta pemrosesan berbasis GCS memungkinkan penggunaan YOLO dengan performa tinggi tanpa membebani UAV.
==================================================================================================================================
Forest fires have a widespread impact on ecosystems, health, and the economy. This study developed a detection system based on Unmanned Aerial Vehicles (UAVs) using image processing object detection with YOLOv8s to improve the effectiveness of fire detection. This system uses a NOIR camera with an RGB approach that is sensitive to the NIR spectrum on the red (R) channel to capture heat emissions from fire objects. Images from the UAV are transmitted in real-time via the internet to the Ground Control Station (GCS) for processing. The study results show that the system can detect fires as small as 0.6 x 0.6 square meters up to 90 meters flying altitude in all variations of hotspot locations, although the YOLOv8s bounding box tends to be larger than the actual fire size due to flame tongue factors and video resolution limitations. Fire area measurements using the Ground Sample Distance (GSD) method show an error range of 0.13–1.71 meters at heights of 60–90 meters across all detection positions, which is still acceptable within the context of aerial object measurement based on the ASPRS 2024 standard for RMSEH of 0.75–1.5 meters. Geotagging of fire locations has an error range of 2.35–7.85 meters (60 meters) and 2.4–8 meters (90 meters) due to limitations of the UAV GNSS, crosswinds, and offset between the camera and UAV GNSS, but it complies with GPS NAVSTAR standards. The conclusion drawn is that the RGB method with NIR sensitivity on the red channel using the NOIR camera can effectively detect small fires, and GCS-based processing enables the use of YOLO with high performance without overloading the UAV.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	UAV, Spectral Imaging, Object Detection, YOLOv8, Kebakaran Hutan, Deteksi Dini. ============================================================ UAV, Spectral Imaging, Object Detection, YOLOv8, Forest Fire, Early Detection.
Subjects:	Q Science > Q Science (General) > Q325.5 Machine learning. Support vector machines. T Technology > TL Motor vehicles. Aeronautics. Astronautics > TL798.N3 Global Positioning System. T Technology > TR Photography > TR810 Aerial photography U Military Science > UG1242 Drone aircraft--Control systems. (unmanned vehicle)
Divisions:	Faculty of Industrial Technology and Systems Engineering (INDSYS) > Physics Engineering > 30201-(S1) Undergraduate Thesis
Depositing User:	Axcel Nirmaulana Ardiansyah
Date Deposited:	31 Jul 2025 07:14
Last Modified:	31 Jul 2025 07:14
URI:	http://repository.its.ac.id/id/eprint/124148

Actions (login required)

View Item