Perbandingan Metode Deep Learning Dan Object-Based Image Analysis (OBIA) Dalam Pemodelan Footprint Bangunan LoD-2 Untuk Evaluasi Kesesuaian Tinggi Bangunan

Ramadhani, Unzilla (2026) Perbandingan Metode Deep Learning Dan Object-Based Image Analysis (OBIA) Dalam Pemodelan Footprint Bangunan LoD-2 Untuk Evaluasi Kesesuaian Tinggi Bangunan. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5016221088-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (9MB) | Request a copy

Abstract

Pertumbuhan kawasan terbangun di Kota Surabaya, khususnya di Kelurahan Pradah Kalikendal, Kecamatan Dukuh Pakis, memerlukan pemantauan spasial yang mampu mempresentasikan kondisi bangunan secara horizontal maupun vertikal. Penelitian ini dibuat untuk membandingkan performa metode Deep Learning berbasis Mask R-CNN dan Object-Based Image Analysis (OBIA) dalam ekstraksi footprint bangunanmodel kota 3D Level of Detail 2 (LoD-2), serta mengevaluasi kesesuaian tinggi bangunan terhadap ketentuan Rencana Detail Tata Ruang (RDTR) Kota Surabaya. Data yang digunakan meliputi data ortofoto, Digital Surface Model (DSM), Digital Terrain Model (DTM), dan data RDTR. Data DSM dan DTM di olah menjadi normalized Digital Surface Model (nDSM) sebagai sumber informasi tinggi bangunan. Hasil uji akurasi menunjukkan bahwa OBIA memiliki performa lebih baik dibandingkan Mask R-CNN, dengan precision 99,05%, recall 99,52%, F1-score 99,28%, komisi 0,95%, dan omisi 0,48%. Sementara itu, Mask R-CNN menghasilkan precision 100%, recall 93,94%, F1-score 96,88%, komisi 0%, dan omisi 6,06%. Model kota 3D LoD-2 berhasil dibangun untuk 6.330 objek bangunan. Validasi tinggi bangunan dengan menggunakan 40 sampel di lapangan menghasilkan RMSE sebesar 0,948 m. Hasil evaluasi RDTR menunjukkan bahwa terdapat 5.458 bangunan atau 86,22% dinyatakan sesuai dengan peraturan RDTR, sedangkan 872 bangunan atau 13,78% dinyatakan tidak sesuai. Peta 3D LoD-2 kesesuaian tinggi bangunan dapat digunakan sebagai instrumen pendukung dalam pengawasan dan pengendalian pemanfaatan ruang.
=================================================================================================================================
The growth of built-up areas in the city of Surabaya, especially in Pradah Kalikendal Village, Dukuh District, requires spatial monitoring that is able to present the condition of the building horizontally and vertically. This study was made to compare the performance of Mask R-CNN-based Deep Learning methods and Object-Based Image Analysis (OBIA) in the extraction of building footprints of 3D Level of Detail 2 (LoD-2) city models, as well as to evaluate the high suitability of buildings to the provisions of the Surabaya City Detailed Spatial Plan (RDTR). The data used included orthophoto data, Digital Surface Model (DSM), Digital Terrain Model (DTM), and RDTR data. DSM and DTM data are processed into normalized Digital Surface Model (nDSM) as a source of high building information. The results of the accuracy test showed that OBIA performed better than R-CNN Mask, with an accuracy of 99.05%, a recall of 99.52%, an F1-score of 99.28%, a commission of 0.95%, and an omission of 0.48%. Meanwhile, the R-CNN Mask produces 100% accuracy, 93.94% recall, 96.88% F1-score, 0% commission, and 6.06% omission. The LoD-2 3D city model was successfully built for 6,330 building objects. Validation of the height of the building using 40 samples in the field resulted in an RMSE of 0.948 m. The results of the RDTR evaluation showed that there were 5,458 buildings or 86.22% declared in accordance with RDTR regulations, while 872 buildings or 13.78% were declared non-compliant. The 3D LoD-2 map of the height of the building can be used as a supporting instrument in the supervision and control of space utilization.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Deep Learning, LiDAR, LoD-2, Mask R-CNN, OBIA, Ortofoto, RDTR Kota Surabaya, Deep Learning, LiDAR, LoD-2, Mask R-CNN, OBIA, Ortophoto, RDTR Surabaya City.
Subjects:	T Technology > TA Engineering (General). Civil engineering (General) > TA590 Topographical surveying T Technology > TR Photography > TR810 Aerial photography
Divisions:	Faculty of Civil Engineering and Planning > Geomatics Engineering > 29202-(S1) Undergraduate Thesis
Depositing User:	Unzilla Ramadhani
Date Deposited:	27 Jul 2026 00:47
Last Modified:	27 Jul 2026 00:47
URI:	http://repository.its.ac.id/id/eprint/137329

Actions (login required)

View Item