Transformer-based Flood Scene Segmentation for Developing Countries Ahan M R

2025-05-06 0 0 819.75KB 7 页 10玖币

侵权投诉

Transformer-based Flood Scene Segmentation for

Developing Countries

Ahan M R∗

BITS Pilani Goa Campus

ahanmr98@gmail.com

Roshan Roy*

BITS Pilani

rroshanroy@gmail.com

Shreyas Sunil Kulkarni

BITS Pilani Hyderabad Campus

sskshreyas@gmail.com

Vaibhav Soni

MANIT Bhopal

vaibsoni@gmail.com

Ashish Chittora

BITS Pilani Goa Campus

ashishc@goa.bits-pilani.ac.in

Abstract

Floods are large-scale natural disasters that often induce a massive number of

deaths, extensive material damage, and economic turmoil. The effects are more

extensive and longer-lasting in high-population and low-resource developing coun-

tries. Early Warning Systems (EWS) constantly assess water levels and other

factors to forecast ﬂoods, to help minimize damage. Post-disaster, disaster response

teams undertake a Post Disaster Needs Assessment (PDSA) to assess structural

damage and determine optimal strategies to respond to highly affected neighbor-

hoods. However, even today in developing countries, EWS and PDSA analysis of

large volumes of image and video data is largely a manual process undertaken by

ﬁrst responders and volunteers. We propose FloodTransformer, which to the best

of our knowledge, is the ﬁrst visual transformer-based model to detect and segment

ﬂooded areas from aerial images at disaster sites. We also propose a custom metric,

Flood Capacity (FC) to measure the spatial extent of water coverage and quantify

the segmented ﬂooded area for EWS and PDSA analyses. We use the SWOC Flood

segmentation dataset and achieve 0.93 mIoU, outperforming all other methods.

We further show the robustness of this approach by validating across unseen ﬂood

images from other ﬂood data sources.

1 Introduction and Context

The Center for Research on the Epidemiology of Disasters, in afﬁliation with the World Health

Organization (WHO), reported that natural disasters accounted for 1.3 million deaths and over USD

2 trillion in economic damage — all between 1998 and 2017 [

]. Flooding related damage is

a factor in most of them [

] and frequent the list of most expensive disasters [

]. Developing

economies of Asia are disproportionately affected and are the worst-hit by ﬂoods, accounting for

44% of all ﬂood disasters from 1987-1997 [

]. India alone registers 1/5th of global deaths from

ﬂoods [

]. Rapid urbanization, global climate change, and rising sea water levels will expose 1.47

billion more people to ﬂood risk, with 89% of them living in low-middle income countries [

]. Flood

Segmentation technology is instrumental for Disaster Prediction and Response is critical to save lives

and livelihoods.

Flood Response

: Typically, disaster management teams complete a Post Disaster Needs Assessment

(PDSA) and rapidly develop infrastructure based on this report on the collected data [

]. Unmanned

Aerial Vehicles (UAVs) are deployed to collect large volumes of image and video data in affected

∗Equal Contribution

35th Conference on Neural Information Processing Systems (NeurIPS 2021), virtual.

arXiv:2210.04218v1 [cs.CV] 9 Oct 2022

regions. PDSA is essential for identiﬁcation of submerged regions, sanity check of large building

structures, debris identiﬁcation, and search-and-rescue (S&R) operations.

Flood Forecasting

: Flood segmentation techniques can be critically important for ﬂooding-related

Early Warning Systems (EWS). According to research in [

], Indians given a ﬂood warning are

twice as likely to evacuate safely than Indians without any notice. which require constant monitoring

of river or sea water levels. Comparison of current levels with historical evidence of ﬂood-prone

water levels can help understand when to trigger warnings appropriately.

Constraints

: Developing countries are plagued by resource and economic constraints. Failure of

macro- and micro- infrastructure planning in Nicaragua led to re-construction on top of an earthquake

faultline [

]. Weak social safety and insurance policies inﬂate recovery time [

]. Economic

vulnerability renders countries like Haiti, Ethiopia, Nepal, El Salvador in a near-permanent state of

emergency alert [

]. In these countries, processing and analysis of large-scale visual data from UAVs

for PDSA in Flood response is a manual process that requires multi-team intervention, which poses a

serious bottleneck in search-and-response speed. Deployment of EWSs is infeasible because human

monitoring of video feeds is too cumbersome and expensive.

AI Technology

: To reduce the burden of manual analysis on crisis responders, Deep learning is

well-suited to scale, automate and expedite these operations. The last few years have witnessed

a tremendous rise in CNN-based image classiﬁcation and segmentation research [

]. However,

CNNs suffer from a well-known problem — large inductive biases. Conceretely, CNNs assume

locality and translation equivariance, which hurt the interpretability of pure CNN-based algorithms.

Recently, visual transformers have garnered attention for image classiﬁcation, segmentation and

object detection tasks [2, 20, 26, 3] for challenging these assumptions with comparable accuracy.

Contributions

: In this work, we propose a hybrid fused CNN-Transformer: FloodTransformer to

tackle ﬂood water segmentation on the Water Segmentation Open Collection (WSOC) dataset [

First, we achieve state-of-the-art results and are the ﬁrst work (to the best of our knowledge) to apply

new transformer-driven research to the ﬂood data domain. Second, our approach is extendable —

we demonstrate the ability of our model to generalize well on unseen data sources. Further local

calibration, if required at all, simply requires weight ﬁne-tuning with previous, region-speciﬁc, ﬂood

scene data. Third, our model does not suffer from data scarcity - it only requires image data input and

not complex sensor data which is hard to collect [

]. Last, the transformer-based encoder applies

recent DL innovations to the ﬂood data domain. Although the hybrid method still uses CNNs in

the decoder network, the aforementioned spatial inductive biases no longer occur throughout the

entire network. Dependencies between patch embeddings are learnt from scratch. This improves the

robustness of our approach.

2 Methodology

To achieve the Flood Scene Understanding, we introduce a Deep Learning model for Flood image

segmentation and quantify the impact of ﬂooding with a custom metric called Flooding Capacity.

2.1 Method

Inspired by Zhang et al. [

], we propose FloodTransformer to solve segmentation for the ﬂood data

domain. It is a fusion architecture of Visual Transformer [

] and Convolution Neural Networks

(CNNs) and its model architecture is displayed in Figure 1.

Complex ﬂooding imagery may contain heterogeneous objects, ﬂooding patterns and backgrounds.

Using the self-attention module of the visual Transformer module from [

] and global vector

representation learned from the CNN network, FloodTransformer fuses the trained embeddings to

learn long-term spatial relationships between the aforementioned entities in images of ﬂood affected

areas. Using Hadamard bilinear product [

], the fusion module fuses information via embeddings

from both parallel streams into a dense representation. The combination of multi-level fusion maps

generates the segmentation output of the model. We summarize each component below, per Zhang et

al. [23].

Transformer Module

: We use the encoder-decoder network using Visual Transformer [

]. The

input image

x∈RH×W×3

is sliced into N patches, where

N=H

F×W

and F is usually set to 16 or

文档加载中……请稍候！
如果长时间未打开，您也可以点击刷新试试。

下载文档到电脑，查找使用更方便

10 玖币 0人已下载

立即下载

摘要：

Transformer-basedFloodSceneSegmentationforDevelopingCountriesAhanMRBITSPilaniGoaCampusahanmr98@gmail.comRoshanRoy*BITSPilanirroshanroy@gmail.comShreyasSunilKulkarniBITSPilaniHyderabadCampussskshreyas@gmail.comVaibhavSoniMANITBhopalvaibsoni@gmail.comAshishChittoraBITSPilaniGoaCampusashishc@goa.bits-...

展开>> 收起<<

Transformer-based Flood Scene Segmentation for Developing Countries Ahan M R.pdf

共7页,预览2页

还剩页未读，继续阅读

声明：本站为文档C2C交易模式，即用户上传的文档直接被用户下载，本站只是中间服务平台，本站所有文档下载所得的收益归上传人(含作者)所有。玖贝云文库仅提供信息存储空间，仅对用户上传内容的表现方式做保护处理，对上载内容本身不做任何修改或编辑。若文档所含内容侵犯了您的版权或隐私，请立即通知玖贝云文库，我们立即给予删除！

Transformer-based Flood Scene Segmentation for Developing Countries Ahan M R

相关推荐

开通VIP享超值会员特权

作者详情

相关内容

热门标签

举报选择: