Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Superpixel-Based Building Damage Detection from Post-earthquake Imagery Using Deep Neural Networks

Jun Wang

TL;DR

The paper tackles rapid building damage assessment after earthquakes by leveraging post-event very high-resolution imagery. It introduces a texture-enhanced, superpixel-based pipeline that couples modified FSAM-based over-segmentation (with spectral angle mapping), Region Adjacency Graph–driven region merging, and a pre-trained SDAE-DNN classifier to learn hierarchical damage-related features. Evaluated on WorldView-2 data from the 2015 Nepal earthquake, the approach achieves strong cross-validated performance (e.g., CV accuracy $0.877$ and Kappa $0.934$) and outperforms conventional classifiers like SVM, RF, ELM, and MLP, demonstrating improved damage detection while reducing manual feature engineering. The method holds practical promise for faster, more reliable damage assessments and aiding post-disaster rescue and reconstruction planning.

Abstract

Building damage detection after natural disasters like earthquakes is crucial for initiating effective emergency response actions. Remotely sensed very high spatial resolution (VHR) imagery can provide vital information due to their ability to map the affected buildings with high geometric precision. However, we suffer from suboptimal performances in detecting damaged buildings due to earthquakes. This paper presents a novel superpixel based approach incorporates Deep Neural Networks (DNN) with a modified segmentation method, for more precise building damage detection from VHR imagery. Firstly, a modified Fast Scanning and Adaptive Merging method is extended to create initial over-segmentation. Secondly, the segments are properly merged based on the Region Adjacent Graph (RAG). Thirdly, a pre-trained DNN using Stacked Denoising Auto-Encoders (SDAE-DNN) is presented, to exploit the rich semantic features for building damage detection. Experimental results on a WorldView-2 imagery from Nepal Earthquake of 2015 demonstrate the feasibility and effectiveness of our method, which could boost detection accuracy through learning more intrinsic and discriminative features, which outperforms other methods using alternative classifiers.

Superpixel-Based Building Damage Detection from Post-earthquake Imagery Using Deep Neural Networks

TL;DR

The paper tackles rapid building damage assessment after earthquakes by leveraging post-event very high-resolution imagery. It introduces a texture-enhanced, superpixel-based pipeline that couples modified FSAM-based over-segmentation (with spectral angle mapping), Region Adjacency Graph–driven region merging, and a pre-trained SDAE-DNN classifier to learn hierarchical damage-related features. Evaluated on WorldView-2 data from the 2015 Nepal earthquake, the approach achieves strong cross-validated performance (e.g., CV accuracy and Kappa ) and outperforms conventional classifiers like SVM, RF, ELM, and MLP, demonstrating improved damage detection while reducing manual feature engineering. The method holds practical promise for faster, more reliable damage assessments and aiding post-disaster rescue and reconstruction planning.

Abstract

Building damage detection after natural disasters like earthquakes is crucial for initiating effective emergency response actions. Remotely sensed very high spatial resolution (VHR) imagery can provide vital information due to their ability to map the affected buildings with high geometric precision. However, we suffer from suboptimal performances in detecting damaged buildings due to earthquakes. This paper presents a novel superpixel based approach incorporates Deep Neural Networks (DNN) with a modified segmentation method, for more precise building damage detection from VHR imagery. Firstly, a modified Fast Scanning and Adaptive Merging method is extended to create initial over-segmentation. Secondly, the segments are properly merged based on the Region Adjacent Graph (RAG). Thirdly, a pre-trained DNN using Stacked Denoising Auto-Encoders (SDAE-DNN) is presented, to exploit the rich semantic features for building damage detection. Experimental results on a WorldView-2 imagery from Nepal Earthquake of 2015 demonstrate the feasibility and effectiveness of our method, which could boost detection accuracy through learning more intrinsic and discriminative features, which outperforms other methods using alternative classifiers.
Paper Structure (7 sections, 6 equations, 4 figures, 2 tables)

This paper contains 7 sections, 6 equations, 4 figures, 2 tables.

Table of Contents

  1. Introduction
  2. BUILDING DAMAGE DETECTION WITH DNN
  3. Initial partition using modified FSAM
  4. Region Merging based on RAG
  5. SDAE-DNN pre-train strategy
  6. EXPERIMENTS AND RESULTS
  7. CONCLUSIONS

Figures (4)

  • Figure 1: Flowchart of the proposed building damage detection method.
  • Figure 2: Region merging based on RAG. Left: Image region merge. Right: Corresponding Region Adjacent Graph (RAG).
  • Figure 3: The Denoising Auto-Encoder (DAE) architecture.
  • Figure 4: The ground truth data and comparison of our detection results with four alternative algorithms. (a) The ground truth data, the yellow points indicate the damaged buildings. (b) (f) Detection results of proposed SDAE-DNN, SVM, RF, ELM and MLP. (Red demonstrates the damaged building areas.)