Segmentasi Lahan Perkebunan Kelapa Sawit Berdasarkan Usia Tanam Pada Citra Satelit

Agustin, Soffiana (2021) Segmentasi Lahan Perkebunan Kelapa Sawit Berdasarkan Usia Tanam Pada Citra Satelit. Doctoral thesis, Institut Teknologi Sepuluh Nopember.

[thumbnail of 05111360010005_Disertation.pdf] Text
05111360010005_Disertation.pdf - Accepted Version
Restricted to Repository staff only until 1 October 2023.

Download (11MB) | Request a copy

Abstract

Kandungan minyak dalam buah kelapa sawit bervariasi, antara lain berdasarkan usia tanaman, sehingga monitoring usia tanaman sawit penting dilakukan. Monitoring perkebunan yang dilakukan secara manual kurang efisien serta sulit diterapkan pada perkebunan yang luas. Penggunaan teknologi untuk monitoring usia tanaman secara otomatis atau semi otomatis dilakukan dengan menggunakan pengolahan citra dan sistem penginderaan jauh seperti citra satelit. Citra satelit yang digunakan dalam pertanian dan perkebunan umumnya adalah citra beresolusi tinggi dan multispektral. Dengan citra multispektral dapat dilakukan berbagai analisis seperti deteksi lahan, mendeteksi perbedaan tanaman, dan manajemen perkebunan lainnya. Untuk melakukan segmentasi lahan sawit berdasarkan usia tanaman, penggunaan citra multispektral memiliki kelemahan yaitu resolusi spasial yang lebih rendah dibandingkan citra panchromatic, serta kurang efisiennya pengelolaan citra multispektral untuk deteksi bentuk tanaman yang merupakan dasar penentuan usia tanaman sawit.
Penelitian ini mengusulkan segmentasi lahan perkebunan kelapa sawit berdasarkan usia tanam menggunakan citra satelit Ikonos dengan band pankromatik saja. Dataset yang digunakan dalam penelitian dibagi menjadi tiga kelompok data sesuai dengan kompleksitas tutupan lahan yang menyertainya dan dipartisi dalam blok berukuran 30x30 piksel. Citra akan disegmentasi kedalam empat kelas, yaitu: bukan kelapa sawit (non-sawit), sawit usia muda, dewasa dan tua. Segmentasi dilakukan menggunakan metode-metode Konvensional dan deep learning. Metode Konvensional yang digunakan didasarkan pada fitur tekstur dan fraktal, yaitu: Local Binary Pattern (LBP), Segmentation-based Fractal Texture Analysis (SFTA), Radially Averaged Power Spectrum (RAPSV) dan Wavelets. Pada penelitian ini dikembangkan fitur mean different peak (MDP) dari nilai RAPSV. Pada penelitian ini metode berbasis deep learning dilakukan dalam dua model. Model pertama, CNN digunakan dalam Ekstraksi Fitur dan klasifikasi SVM. Metode kedua, CNN digunakan dalam ekstraksi fitur sekaligus klasifier. Arsitektur yang digunakan dalam penelitian ini adalah AlexNet, VGG-16 dan VGG-19 dengan mengusulkan fine-tuning Model-1 dan Model-2 pada pre-trained network.
Pada metode konvensional, kombinasi fitur terbaik yang diperoleh adalah Wavelet, RAPSV dan MDP dengan mesin pembelajaran Multilayer Perceptron. Akurasi rata-rata tertinggi pada model ini adalah 79,65%. Akurasi rata-rata metode deep learning-based 74,81% dengan Alex-Net fine-tune Model-1 dengan pengklasifikasi SVM dan 84,94% dengan fine-tune Model-2 VGG-19 sebagai ekstraksi fitur sekaligus pengklasifikasi. Dari penelitian yang dilakukan ditemukan bahwa hanya dengan menggunakan citra pankromatik dapat digunakan untuk segmentasi berdasarkan umur tanam sehingga dapat menekan biaya pencitraan satelit yang mahal.
=====================================================================================================
The oil content in oil palm fruit varies, among others, based on the age of the plant, so it is
essential to monitor the age of the oil palm plantations. Plantation monitoring which is done
manually is less efficient and difficult to apply for large plantations. The use of technology for
automatic or semi-automatic monitoring of plant age is carried out by using image processing from
remote sensing systems such as satellite images. Satellite images used in agriculture and plantations
are generally high resolution and multispectral imagery. With this multispectral image, various
analyzes can be carried out such as land detection, detecting plant differences, and other plantation
management. To segment oil palm plantation based on plant age, the use of multispectral imagery
has weaknesses, such as lower spatial resolution than panchromatic images, and less efficient
multispectral image management for detecting plant shapes which is the basis for determining the
age of oil palms.
This study proposes the segmentation of oil palm plantations based on planting age using
Ikonos satellite imagery with only panchromatic bands. The dataset used in the study was divided
into three data groups according to the complexity of the accompanying land cover and partitioned
into blocks measuring 30x30 pixels. The image will be segmented into four classes, namely: non�oil palm (non-oil palm), young, mature and old palm oil. Segmentation is performed using
conventional methods and deep learning. The conventional method used is based on texture and
fractal features, namely: Local Binary Pattern (LBP), Segmentation-based Fractal Texture Analysis
(SFTA), Radially Averaged Power Spectrum (RAPSV) and Wavelets. In this study, the mean
different peak (MDP) feature of the RAPSV value was developed. The deep learning-based method
in this study is performed with two models. The first model, CNN is used in Feature Extraction with
SVM classifier. The second method, CNN is used in feature extraction as well as a classifier. The
architecture used in this research is AlexNet, VGG-16 and VGG-19 by proposing the fine-tuning
Model-1 and Model-2 of the pre-trained network.
In the conventional method, the best combination of features obtained is Wavelet, RAPSV
and MDP with the Multilayer Perceptron learning machine. The highest average accuracy in this
model is 79.65%. The average accuracy of the deep learning-based method is 74.81% with the Alex�Net fine-tune Model-1 with SVM classifier and 84.94% with the VGG-19 fine-tune Model-2 as
feature extraction as well as a classifier. From the research conducted, it was found that only using
panchromatic imagery can be used for segmentation based on planting age so as to reduce the cost
of expensive satellite imagery.

Item Type: Thesis (Doctoral)
Uncontrolled Keywords: Segmentasi, perkebunan kelapa sawit, usia tanam, Ikonos, pankromatik, tekstur, deep learning.
Subjects: S Agriculture > SB Plant culture > SB409.58 Plant propagation. Including in vitro propagation
T Technology > T Technology (General) > T385 Visualization--Technique
Divisions: Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Informatics Engineering > 55001-(S3) PhD Thesis
Depositing User: Soffiana Agustin
Date Deposited: 12 Jan 2022 08:17
Last Modified: 12 Jan 2022 08:17
URI: http://repository.its.ac.id/id/eprint/84000

Actions (login required)

View Item View Item