查詢結果分析
相關文獻
- 新舊影像分割技術於衛星影像萃取建物輪廓之比較
- 應用Watershed影像分割法於高解析QuickBird衛星影像進行道路區塊之萃取
- 應用改良式Watershed技術進行高解析衛星影像影像分割
- 使用融合的RDB U-Nets分割衛星影像中的水禽禽場
- Form Segmentation and Component Classification for Clinic Document Image Analysis
- 遙測衛星影像估測玉山事業區大水窟森林火災之研究
- 空間資訊及土壤沖蝕模式之整合應用
- 四元樹法辨釋衛星影像之研究
- Visualization and 3D Reconstruction for Ultrasound Images of Carotid Arteries with Plaques
- 航照、衛星影像應用於臺北盆地土地利用分類及變遷分析
頁籤選單縮合
| 題 名 | 新舊影像分割技術於衛星影像萃取建物輪廓之比較=Comparison of New and Traditional Image Segmentation Techniques for Extracting Building Footprints from Satellite Imagery |
|---|---|
| 作 者 | 林恩楷; 林政緯; 邵國士; 高郡汝; | 書刊名 | 中興工程 |
| 卷 期 | 168 2025.07[民114.07] |
| 頁 次 | 頁(E)1-(E)10 |
| 分類號 | 440.984 |
| 關鍵詞 | 影像分割; 衛星影像; 建物輪廓; Image segmentation; Satellite imagery; Building footprints; Mask R-CNN; |
| 語 文 | 中文(Chinese) |
| 中文摘要 | 利用影像分割概念於遙測影像中萃取各式特徵物已行之有年,然而近年深度學習(Deep Learning)發展快 速,已為影像分割領域帶來截然不同的新面貌。本研究以高解析度衛星影像中的建物輪廓為萃取標的,比較 2 種 分別於過去、現在之主流影像分割技術,分別是過去常用之多尺度分割法(Multiresolution Segmentation)、及 以現今常見的深度學習模型 Mask R-CNN,以探討 2 種影像分割技術對於 LOD-1 建物輪廓萃取的差異。 成果顯示,使用後起之秀—深度學習模型 Mask R-CNN 所萃取的 LOD-1 建物輪廓較多尺度分割法成果佳, 且差異顯著,並根據此研究成果,實際運用Mask R-CNN於寮國—永珍、越南—胡志明市、柬埔寨—金邊、緬甸— 仰光都會區,共 1,400 平方公里之建物輪廓萃取。 |
| 英文摘要 | The concept of using image segmentation to extract various features from remote sensing imagery has been in practice for many years. However, with the rapid development of deep learning in recent years, it has brought a completely new perspective to the field of image segmentation. This study focuses on extracting building footprints from high-resolution satellite images and compares two mainstream image segmentation techniques, one from the past and one from the present. These techniques are the traditional multiresolution segmentation method, commonly used in the past, and the current deep learning model, Mask R-CNN, to investigate the differences between the two methods in extracting LOD-1 building footprints. The results show that the deep learning model, Mask R-CNN, which is the newer approach, produces better results in extracting LOD-1 building footprints compared to the multiresolution segmentation method, with significant differences. Based on these findings, Mask R-CNN was practically applied to extract building footprints in metropolitan areas across 1,400 square kilometers, including Vientiane (Laos), Ho Chi Minh City (Vietnam), Phnom Penh (Cambodia), and Yangon (Myanmar). |
本系統中英文摘要資訊取自各篇刊載內容。