Deteksi Perubahan pada Citra Foto Udara Bi-temporal Menggunakan Spatial-Temporal Attention Neural Network dan Neighborhood K-NN Second Stage Classification

Al-Islami, Muhammad Izzuddin (2021) Deteksi Perubahan pada Citra Foto Udara Bi-temporal Menggunakan Spatial-Temporal Attention Neural Network dan Neighborhood K-NN Second Stage Classification. Undergraduate thesis, Institut Teknologi Sepuluh Nopember.

Text 05111740000035-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only until 1 October 2023. Download (3MB) | Request a copy

Abstract

Deteksi perubahan (change detection) adalah proses mengidentifikasi perbedaan pada sebuah keadaan dari sebuah benda dengan melakukan pemantauan dalam waktu yang berbeda. Deteksi perubahan pada citra penginderaan jarak jauh berperan penting pada perubahan penggunaan lahan dan land cover, pengawasan perubahan pada hutan dan lahan hijau, pengawasan ekosistem, penelitian perkembangan kawasan urban, manajemen sumber daya dan penilaian dampak kerusakan dikarenakan bencana. Karena itu, dibutuhkan sebuah sistem untuk mendeteksi perubahan pada citra penginderaan jarak jauh yang akurat dan handal.
Untuk membuat sebuah sistem pendeteksi perubahan pada citra penginderaan jarak jauh yang akurat dan handal, maka didesain sebuah sistem deteksi perubahan berbasis jaringan deep learning Spatial-Temporal Attention Neural Network untuk menghasilkan output berupa label kelas berubah/tidak berubah dari feature distance kedua citra foto udara pada setiap piksel. K-NN classifier akan diterapkan untuk mengubah nilai piksel sesuai dengan label piksel tetangganya dengan aturan majority vote.
Hasil eksperimen menunjukkan nilai F1 score tertinggi untuk hasil deteksi perubahan dicapai ketika menggunakan optimizer Adagrad dengan learning rate 0.0001, dan nilai k=7 untuk besar window 3x3 pada post-proses. Hasil dari evaluasi pada model yang dibangun menghasilkan precision sebesar 77,456%, recall sebesar 87,232%, F1-score sebesar 82,054%.
======================================================================================================
Change Detection is the process of identifying differences in a state of an object by monitoring in different times. Detection of changes in remote sensing plays an important role in changes in land use and land cover, supervision of changes in forests and green land, ecosystem supervision, development of the development of urban areas, resource management and assessment of damage impact due to disasters. Therefore, a system is needed to detect changes in accurate and reliable way.
To create an accurate and reliable change detection system, we designed a system using deep learning Spatial-Temporal Attention Neural Network to produce output in the form of a change / unchanged class label from the feature distance from both aerial image on every pixel. K-NN Classifier will then be applied to change the pixel value according to the neighbor's pixel label based on the Rules of Majority Vote.
The experiments result show the value of the highest F1 score for the changes detection is achieved when using the Adagrad optimizer with Learning Rate of 0.0001, and the value of K = 7 for 3x3 window in the post-process. The results of the evaluation in the model built produced a precision of 77.456%, recall of 87.232%, F1-score of 82.054%.

Item Type:	Thesis (Undergraduate)
Uncontrolled Keywords:	Deteksi Perubahan, citra foto udara, pengolahan citra, deep learning, Change Detection, Aerial Imagery, Image Processing
Subjects:	Q Science > QA Mathematics > QA76.87 Neural networks (Computer Science) R Medicine > R Medicine (General) > R858 Deep Learning T Technology > TA Engineering (General). Civil engineering (General) > TA1637 Image processing--Digital techniques
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Informatics Engineering > 55201-(S1) Undergraduate Thesis
Depositing User:	Muhammad Izzuddin Al-Islami
Date Deposited:	07 Aug 2021 03:38
Last Modified:	07 Aug 2021 03:38
URI:	http://repository.its.ac.id/id/eprint/85046

Actions (login required)

View Item