Multi-task learning based post-disaster mapping via multi-modal remote sensing observations

• Project/Area Number: 23K13419
• Research Category: Grant-in-Aid for Early-Career Scientists
• Allocation Type: Multi-year Fund
• Review Section: Basic Section 22050:Civil engineering plan and transportation engineering-related
• Research Institution: The University of Tokyo
• Principal Investigator: 袁 巍 東京大学, 空間情報科学研究センター, 客員研究員 (60837475)
• Project Period (FY): 2023-04-01 – 2026-03-31
• Project Status: Granted (Fiscal Year 2023)
• Budget Amount: ¥4,550,000 (Direct Cost: ¥3,500,000、Indirect Cost: ¥1,050,000); Fiscal Year 2025: ¥910,000 (Direct Cost: ¥700,000、Indirect Cost: ¥210,000); Fiscal Year 2024: ¥1,690,000 (Direct Cost: ¥1,300,000、Indirect Cost: ¥390,000); Fiscal Year 2023: ¥1,950,000 (Direct Cost: ¥1,500,000、Indirect Cost: ¥450,000)
• Keywords: Photogrammetry / Remote Sensing / Change Detection / Data Fusion / Disaster assesment

Outline of Research at the Start

Automatic post-disaster damage mapping is important, especially in Japan where natural disasters cause heavy losses frequently. Focusing on the perspective of physical damage mapping, this research proposed to achieve rapid, accurate, and cost-effective post-disaster damage mapping based on multi-modal remote sensing observations (and deep learning methods. There are three main steps in the proposed methods: 1) Multi-modal remote sensing data fusion for feature extraction; 2) Progressive-supervision for multi-task urban mapping; 3) transfer learning based post-disaster damage mapping.