Analisis Tree Counting Tanaman Kelapa Sawit Menggunakan Algoritma Template Matching Dan Deep Learning Mask R-CNN

Ristawan, Sulthan Hafizh (2024) Analisis Tree Counting Tanaman Kelapa Sawit Menggunakan Algoritma Template Matching Dan Deep Learning Mask R-CNN. Other thesis, Institut Teknologi Sepuluh Nopember.

	Text Undergraduate Thesis_Sulthan Hafizh Ristawan.pdf - Accepted Version Restricted to Repository staff only until 1 October 2026. Download (14MB) | Request a copy
	Text 5016201109-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only until 1 October 2026. Download (14MB) | Request a copy

Abstract

Indonesia merupakan salah satu produsen dan eksportir minyak sawit terbesar di dunia. Menurut Badan Pusat Statistik (BPS) pada tahun 2022 produksi minyak sawit mengalami peningkatan menjadi 46,82 juta ton. Kelapa sawit memiliki potensi yang besar, sehingga diperlukan teknologi untuk menganalisis produktivitasnya. Maka dari itu, penghitungan pohon kelapa sawit menjadi aspek penting dalam pemantauan lahan, perawatan tanaman, dan perencanaan produksi yang efisien. Namun, penghitungan pohon kelapa sawit secara manual membutuhkan waktu, tenaga kerja yang besar, dan cenderung kurang efisien di perkebunan yang luas. Selain itu, kurangnya akurasi dalam estimasi jumlah pohon dan umur tanaman dapat mempengaruhi perencanaan dan manajemen lahan. Pengembangan algoritma pada keilmuan geomatika dapat melakukan penghitungan pohon kelapa sawit secara otomatis melalui data foto udara yang dikombinasikan dengan metode template matching dan deep learning yang diharapkan dapat memberikan solusi yang efisien dan akurat. Dalam pengolahannya diambil training sample sebanyak 620 sampel pohon kelapa sawit. Setelah dilakukan pengolahan, didapatkan hasil interpretasi visual yakni didapatkan sebanyak 6.195 pohon, selanjutnya dengan menggunakan metode deep learning mask R-CNN sebanyak 6.359 pohon, serta menggunakan metode template matching sebanyak 6.756 pohon. Melalui uji akurasi menggunakan metode confussion matrix dengan mereferensi dari interpretasi visual dibandingkan dengan metode template matching dan deep learning maka didapatkan nilai overall accuracy menggunakan metode deep learning mask R-CNN yakni 94,99% dan nilai overall accuracy menggunakan metode template matching sebesar 86,77%. Berdasarkan pengolahan yang sudah dilakukan tidak semua titik pohon berada pada posisi yang tepat, ketidaktepatan ini didapatkan karena berbagai faktor penyebab seperti kondisi ortofoto yang mempunyai perbedaan warna gelap dan terang, visual pohon kelapa sawit yang kurang bagus, kondisi rapat renggangnya pohon kelapa sawit serta kondisi kelapa sawit dalam ortofoto yang lumayan acak.
=================================================================================================================================
Indonesia is one of the largest producers and exporters of palm oil in the world. According to the Central Bureau of Statistics (BPS) in 2022 palm oil production has increased to 46.82 million tons. Palm oil has great potential, so technology is needed to analyze its productivity. Therefore, palm oil tree counting is an important aspect in land monitoring, plant maintenance, and efficient production planning. However, manual palm tree counting is time-consuming, labor-intensive, and tends to be less efficient in large plantations. In addition, the lack of accuracy in estimating the number of trees and age of the crop can affect land planning and management. Algorithm development in geomatics science can perform automatic palm tree counting through aerial photo data combined with template matching and deep learning methods that are expected to provide efficient and accurate solutions. In the processing, 620 training samples of oil palm trees were taken. After processing, the results of visual interpretation were obtained as many as 6.195 trees, then using the deep learning mask R-CNN method as many as 6.359 trees, and using the template matching method as many as 6.756 trees. Through the accuracy test using the confusion matrix method with reference to visual interpretation compared to the template matching and deep learning methods, the overall accuracy value using the deep learning mask R-CNN method is 94,99% and the overall accuracy value using the template matching method is 86,77%. Based on the processing that has been done, not all tree points are in the right position, this inaccuracy is obtained due to various causal factors such as orthophoto conditions that have differences in dark and light colors, visual palm trees that are not good, the condition of the tightness of the palm trees and the condition of the palm in the orthophoto which is quite random.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Tree Counting, Oil Palm Tree, Template Matching, Deep Learning, Deep Learning, Kelapa Sawit
Subjects:	G Geography. Anthropology. Recreation > G Geography (General) G Geography. Anthropology. Recreation > G Geography (General) > G70.212 ArcGIS. Geographic information systems. G Geography. Anthropology. Recreation > G Geography (General) > G70.5.I4 Remote sensing
Divisions:	Faculty of Civil, Planning, and Geo Engineering (CIVPLAN) > Geomatics Engineering > 29202-(S1) Undergraduate Thesis
Depositing User:	Sulthan Hafizh Ristawan
Date Deposited:	03 Sep 2024 08:44
Last Modified:	03 Sep 2024 08:44
URI:	http://repository.its.ac.id/id/eprint/109798

Actions (login required)

View Item