Segmentasi Jalan Daerah Nonperkotaan pada Peta Dasar Ortho Badan Pertanahan Nasional Menggunakan DeepLabV3+ dengan Modul Dense Depthwise Dilated Separable Spatial Pyramid Pooling

Ramadhani, Raihan (2025) Segmentasi Jalan Daerah Nonperkotaan pada Peta Dasar Ortho Badan Pertanahan Nasional Menggunakan DeepLabV3+ dengan Modul Dense Depthwise Dilated Separable Spatial Pyramid Pooling. Bachelor thesis, Institut Teknologi Kalimantan.

[img] Text
11211069_cover.pdf
Restricted to Registered users only until 4 October 2027.
Download (314kB) | Request a copy
[img] Text
11211069_statement_of_authenticity.pdf
Restricted to Registered users only until 4 October 2027.
Download (379kB) | Request a copy
[img] Text
11211069_publishing_agreement.pdf
Restricted to Registered users only until 4 October 2027.
Download (321kB) | Request a copy
[img] Text
11211069_approval_sheet.pdf
Restricted to Registered users only until 4 October 2027.
Download (244kB) | Request a copy
[img] Text
11211069_preface.pdf
Restricted to Registered users only until 4 October 2027.
Download (271kB) | Request a copy
[img] Text
11211069_abstract_id.pdf
Restricted to Registered users only until 4 October 2027.
Download (314kB) | Request a copy
[img] Text
11211069_abstract_en.pdf
Restricted to Repository staff only until 4 October 2027.
Download (315kB) | Request a copy
[img] Text
11211069_table_of_content.pdf
Restricted to Repository staff only until 4 October 2027.
Download (345kB) | Request a copy
[img] Text
11211069_illustrations.pdf
Restricted to Repository staff only until 4 October 2027.
Download (337kB) | Request a copy
[img] Text
11211069_tables.pdf
Restricted to Repository staff only until 4 October 2027.
Download (267kB) | Request a copy
[img] Text
11211069_chapter_1.pdf
Restricted to Repository staff only until 4 October 2027.
Download (334kB) | Request a copy
[img] Text
11211069_chapter_2.pdf
Restricted to Repository staff only until 4 October 2027.
Download (1MB) | Request a copy
[img] Text
11211069_chapter_3.pdf
Restricted to Repository staff only until 4 October 2027.
Download (620kB) | Request a copy
[img] Text
11211069_chapter_4.pdf
Restricted to Repository staff only until 4 October 2027.
Download (1MB) | Request a copy
[img] Text
11211069_conclusions.pdf
Restricted to Repository staff only until 4 October 2027.
Download (311kB) | Request a copy
[img] Text
11211069_bibliography.pdf
Restricted to Registered users only until 4 October 2027.
Download (285kB) | Request a copy
[img] Text
11211069_enclosure.pdf
Restricted to Repository staff only until 4 October 2027.
Download (522kB) | Request a copy
[img] Text
11211069_paper.pdf
Restricted to Repository staff only until 4 October 2027.
Download (954kB) | Request a copy
[img] Text
11211069_presentation.pdf
Restricted to Repository staff only until 4 October 2027.
Download (2MB) | Request a copy
[img] Text
11211069_Form. TA-020.pdf
Restricted to Repository staff only until 4 October 2027.
Download (262kB) | Request a copy
[img] Text
11211069_cover.pdf
Restricted to Registered users only until 4 October 2027.
Download (208kB) | Request a copy
[img] Text
11211069_cover.pdf
Restricted to Registered users only until 4 October 2027.
Download (208kB) | Request a copy

Abstract

Pemetaan jalan yang akurat di daerah nonperkotaan Indonesia menjadi tantangan krusial akibat struktur jalan yang kompleks dan seringkali tidak terdokumentasi dengan baik. Penelitian ini bertujuan untuk (1) menganalisis pengaruh berbagai modul attention (SE-block, scSE, ECA-Net) terhadap kinerja segmentasi jalan pada arsitektur DeepLabV3+ yang dimodifikasi dengan Dense Depthwise Dilated Separable Spatial Pyramid Pooling (DenseDDSSPP), dan (2) membandingkan serta merekomendasikan arsitektur terbaik antara model usulan dengan DeepLabV3+ standar (ASPP) dan RFE-LinkNet. Metode penelitian ini adalah eksperimen segmentasi semantik menggunakan arsitektur DeepLabV3+ dengan backbone Xception dan modul DenseDDSSPP pada dataset baru, BackroadsBPNID, yang dikembangkan dari peta dasar Ortho Badan Pertanahan Nasional (BPN). Evaluasi dilakukan berdasarkan metrik Precision, F1-Score, Intersection over Union (IoU), serta jumlah parameter untuk efisiensi komputasi. Hasil penelitian menunjukkan bahwa penambahan modul attention scSE memberikan peningkatan performa paling optimal, dengan F1-Score tertinggi (0,7553) dan IoU (0,6316). Pada perbandingan arsitektur, model usulan (DeepLabV3+ dengan DenseDDSSPP + scSE) terbukti paling unggul, melampaui DeepLabV3+ standar (IoU 0,6289) dan RFE-LinkNet (IoU 0,5168). Selain itu, model ini juga lebih efisien dengan 20,87 juta parameter, lebih sedikit dibandingkan DeepLabV3+ standar (21,64 juta). Arsitektur DeepLabV3+ dengan modul DenseDDSSPP dan attention scSE merupakan konfigurasi terbaik yang menawarkan keseimbangan superior antara akurasi, efisiensi, dan kualitas visual untuk segmentasi jalan nonperkotaan di Indonesia.

Item Type: Thesis (Bachelor)
Subjects: Q Science > QA Mathematics > QA76 Computer software
T Technology > TE Highway engineering. Roads and pavements
Divisions: Jurusan Matematika dan Teknologi Informasi > Informatika
Depositing User: Raihan Ramadhani
Date Deposited: 07 Jul 2025 02:41
Last Modified: 07 Jul 2025 02:41
URI: http://repository.itk.ac.id/id/eprint/22808

Actions (login required)

View Item View Item