Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Graph Attention Convolutional U-NET: A Semantic Segmentation Model for Identifying Flooded Areas

Muhammad Umair Danish, Madhushan Buwaneswaran, Tehara Fonseka, Katarina Grolinger

TL;DR

Addressing flood-area segmentation from aerial imagery, the paper proposes GAC-UNET, a graph-attention enhanced U-Net that combines GATConv, Chebyshev convolutions, and a Center of Mass layer. It explores transfer learning and model reprogramming to cope with limited labeled data, showing that transfer learning yields greater improvements. On the Kaggle Flood Area Segmentation dataset, GAC-UNET achieves Dice 0.94, IoU 0.89, and mAP 0.91, outperforming CNN/U-Net, ENet, and SegFormer baselines. The work highlights the practical potential of graph-based architectures for accurate, pixel-level flood mapping and lays groundwork for real-time disaster response and broader environmental segmentation tasks.

Abstract

The increasing impact of human-induced climate change and unplanned urban constructions has increased flooding incidents in recent years. Accurate identification of flooded areas is crucial for effective disaster management and urban planning. While few works have utilized convolutional neural networks and transformer-based semantic segmentation techniques for identifying flooded areas from aerial footage, recent developments in graph neural networks have created improvement opportunities. This paper proposes an innovative approach, the Graph Attention Convolutional U-NET (GAC-UNET) model, based on graph neural networks for automated identification of flooded areas. The model incorporates a graph attention mechanism and Chebyshev layers into the U-Net architecture. Furthermore, this paper explores the applicability of transfer learning and model reprogramming to enhance the accuracy of flood area segmentation models. Empirical results demonstrate that the proposed GAC-UNET model, outperforms other approaches with 91\% mAP, 94\% dice score, and 89\% IoU, providing valuable insights for informed decision-making and better planning of future infrastructures in flood-prone areas.

Graph Attention Convolutional U-NET: A Semantic Segmentation Model for Identifying Flooded Areas

TL;DR

Addressing flood-area segmentation from aerial imagery, the paper proposes GAC-UNET, a graph-attention enhanced U-Net that combines GATConv, Chebyshev convolutions, and a Center of Mass layer. It explores transfer learning and model reprogramming to cope with limited labeled data, showing that transfer learning yields greater improvements. On the Kaggle Flood Area Segmentation dataset, GAC-UNET achieves Dice 0.94, IoU 0.89, and mAP 0.91, outperforming CNN/U-Net, ENet, and SegFormer baselines. The work highlights the practical potential of graph-based architectures for accurate, pixel-level flood mapping and lays groundwork for real-time disaster response and broader environmental segmentation tasks.

Abstract

The increasing impact of human-induced climate change and unplanned urban constructions has increased flooding incidents in recent years. Accurate identification of flooded areas is crucial for effective disaster management and urban planning. While few works have utilized convolutional neural networks and transformer-based semantic segmentation techniques for identifying flooded areas from aerial footage, recent developments in graph neural networks have created improvement opportunities. This paper proposes an innovative approach, the Graph Attention Convolutional U-NET (GAC-UNET) model, based on graph neural networks for automated identification of flooded areas. The model incorporates a graph attention mechanism and Chebyshev layers into the U-Net architecture. Furthermore, this paper explores the applicability of transfer learning and model reprogramming to enhance the accuracy of flood area segmentation models. Empirical results demonstrate that the proposed GAC-UNET model, outperforms other approaches with 91\% mAP, 94\% dice score, and 89\% IoU, providing valuable insights for informed decision-making and better planning of future infrastructures in flood-prone areas.

Paper Structure

This paper contains 21 sections, 7 equations, 4 figures, 1 table.

Table of Contents

  1. Introduction
  2. Related Work
  3. Background and Preliminaries
  4. U-Net
  5. Efficient Neural Network
  6. SegFormer
  7. Methodology
  8. Graph Attention Convolutional UNET
  9. Transfer learning and model reprogramming
  10. Loss functions
  11. Binary cross-entropy loss
  12. Dice loss
  13. Evaluation and Results
  14. Dataset
  15. Pre-processing
  16. ...and 6 more sections

Figures (4)

  • Figure 1: The proposed Graph Attention Convolutional U-NET (GAC-UNET) architecture consists of a series of convolutional layers followed by pooling layers in its encoder part. Then, the graph attention convolutional, Chebyshev graph convolutional, and center of mass layers are stacked between the encoder and decoder. The decoder also consists of a series of convolutional layers.
  • Figure 2: Sample images and corresponding masks from the dataset
  • Figure 3: Output comparison of traditionally trained SegFormer, SegFormer with transfer learning, and segformer with model reprogramming
  • Figure 4: Output comparison of our proposed model with baselines (U-Net, E-net, SegFormer, and SegFormer with transfer learning)