(For USM Staff/Student Only)

EngLib USM > Ω School of Aerospace Engineering >

A camera-aided flight control & navigation system for uav in sustainable oil palm plantation

A camera-aided flight control & navigation system for uav in sustainable oil palm plantation / Ernest Tay Yue Yang
Malaysia adalah salah satu pengeksport minyak sawit terbesar di dunia dengan jumlah 19 516 141 tan minyak sawit mentah pada tahun lalu, 2019. Ini mencerminkan sumbangan industri kelapa sawit terhadap ekonomi Malaysia. Banyak teknologi telah diusulkan untuk diterapkan dalam industri dan banyak yang menunjukkan hasil yang menjanjikan, tetapi masih belum ada solusi yang mampu secara berautonomi mengesan dan menyembur racun perosak pada pokok kelapa sawit dalam lading kelapa sawit. Oleh itu, dalam penyelidikan ini, penyelesaian yang menggunakan pembelajaran mendalam telah dikaji untuk membolehkan drone mengesan pokok kelapa sawit dengan tepat. Penyelidikan ini bertujuan untuk membandingkan prestasi tiga kaedah pengesanan objek yang terkenal bernama Faster R-CNN, SSD dan juga YOLO untuk memahami prestasi setiap kaedah tersebut dalam pengesanan pokok kelapa sawit dalam lading kelapa sawit. Kaedah terbaik kemudian dipilih berdasarkan skor ketepatan purata di mana modal pengesanan objek, Faster R-CNN telah mencapai skor ketepatan purata setinggi 0.972 untuk ambang 0.5, YOLOv3 pula mencapai 0.931 dan SSD mencapai 0.383. Beberapa gambar pokok kelapa sawit kemudian digunakan untuk menguji kecekapan kaedah pengesanan pokok kelapa sawit secara manual. Kaedah yang dipilih kemudian akan disatukan dengan perkakasan sistem tertanam, NVIDIA Jetson Nano. Kaedah perbandingan yang dicadangkan menunjukkan hasil yang tepat jika dibandingkan dengan hasil ujian pemerhatian manual mengesan kotak pengikat menggunakan mata. Akhirnya, modal pengesanan objeck YOLOv3 dipilih sebagai modal yang akan di muat dalam kepada NVIDIA Jetson Nano kerana YOLOv3 mempunyai vi kelajuan pengesanan object yang tinggi serta menunjukkan ketepatan pengesanan objek dalam gambar berlainan kecerahan yang tinggi. _______________________________________________________________________________________________________ Malaysia is one of the largest exporters of palm oil in the world boasting a sum of 19 516 141 Mg of crude palm oil just last year, 2019. This reflects the contribution of the oil palm industry towards Malaysia's economy. Many technologies have been proposed to be applied within the industry, and many showed promising results, but there are still no solutions that are capable of autonomously detect and apply pesticides on oil palm trees within the oil plam plantation. Therefore, in this research, a solution that utilised deep learning has been studied to allow drones to accurately detect oil palm trees. This research compares the performance of three well-known state of the art object detection methods to understand each of the object detection models' performance in detecting oil palm trees. The best object detection model is then selected based on its average precision scores where Faster R-CNN object detection model achieved an average precision of 0.972 for the threshold of 0.5, YOLOv3 on the other hand achieved 0.931 and SSD achieved 0.383. Random images of palm tree crowns will then be used to test the reliability of the model manually. The selected object detetion model will then be integrated with embedded system hardware, the NVIDIA Jetson Nano. The proposed method of the comparison showed accurate results when compared to the results from manual observation tests by observing bounding boxes within the image using the human eye. Lastly, YOLOv3 object detection model was chosen to be integrated with the NVIDIA Jetson Nano because it is quick to detect objects and also shows high accuracy and rubostness when detecting object within images of different light intensity.
Contributor(s):
Ernest Tay Yue Yang - Author
Primary Item Type:
Final Year Project
Identifiers:
Accession Number : 875007945
Language:
English
Subject Keywords:
palm oil; technologies; pesticides
First presented to the public:
8/1/2020
Original Publication Date:
10/2/2020
Previously Published By:
Universiti Sains Malaysia
Place Of Publication:
School of Aerospace Engineering
Citation:
Extents:
Number of Pages - 83
License Grantor / Date Granted:
  / ( View License )
Date Deposited
2020-10-02 17:14:32.217
Submitter:
Mohd Jasnizam Mohd Salleh

All Versions

Thumbnail Name Version Created Date
A camera-aided flight control & navigation system for uav in sustainable oil palm plantation1 2020-10-02 17:14:32.217