SCHOOL OF COMPUTER SCIENCE AND ENGINEERING

Mini Project Report

On

A Deep Learning Model For Post-Disaster Flood Mapping In SAR Images

submitted in partial fulfillment of the requirements for the award of the degree of

Bachelor of Engineering

IN

COMPUTER SCIENCE AND ENGINEERING

Submitted By
Name USN
Sinhcan H 01FE22BCS055
Varda I Pattanshetty 01FE22BCS189
Varsha I Pattanshetty 01FE22BCS191
Sujal Naduvinamani 01FE22BCS248

Under the guidance of

Dr. Sumaiya Pathan

School of Computer Science and Engineering

KLE Technological University, Hubballi

2024-2025
 SCHOOL OF COMPUTER SCIENCE AND ENGINEERING

2024-25

CERTIFICATE

This is to certify that project entitled “A Deep Learning Model For Post-Disaster Flood Map-
ping In SAR Images” is a bonafied work carried out by the student team Sinchan H(01FE22BCS055),
Varda I P(01FE22BCS189), Varsha I P(01FE22BCS191), Sujal N(01FE22BCS248), in partial
fulfillment of the completion of 5th semester B. E. course during the year 2024 – 2025. The
project report has been approved as it satisfies the academic requirement with respect to the
project work prescribed for the above said course.

Guide Name Head, SoCSE

Dr. Sumaiya Pathan Dr. Vijayalakshmi.M.

External Viva-Voce

Name of the examiners Signature with date

1 ————————————– ————————–

2 ————————————— ————————–
 ABSTRACT
We would like to thank our faculty and management for their professional guidance towards
the completion of the mini project work. We take this opportunity to thank Dr. Ashok
Shettar, Pro-Chancellor, Dr. P.G Tewari, Vice-Chancellor and Dr. B.S.Anami, Registrar for
their vision and support.
We also take this opportunity to thank Dr. Meena S. M, Professor and Dean of Faculty, SoCSE
and Dr. Vijayalakshmi M, Professor and Head, SoCSE for having provided us direction and
facilitated for enhancement of skills and academic growth.
We thank our guide Dr.Sumaiya Pathan, Assistant Professor and SoCSE for the constant
guidance during interaction and reviews.
We extend our acknowledgment to the reviewers for critical suggestions and inputs. We also
thank Project coordinator Dr. Uday Kulkarni, and reviewers for their suggestions during the
course of completion. We express gratitude to our beloved parents for constant encouragement
and support.

Keywords : Flood Mapping, Synthetic Aperture Radar(SAR), U-Net, CycleGAN, Disaster

Management, Image Segmentation

i
 ACKNOWLEDGEMENT
We would like to thank our faculty and management for their professional guidance towards
the completion of the mini project work. We take this opportunity to thank Dr. Ashok
Shettar, Pro-Chancellor, Dr. P.G Tewari, Vice-Chancellor and Dr. B.S.Anami, Registrar for
their vision and support.
We also take this opportunity to thank Dr. Meena S. M, Professor and Dean of Faculty, SoCSE
and Dr. Vijayalakshmi M, Professor and Head, SoCSE for having provided us direction and
facilitated for enhancement of skills and academic growth.
We thank our guide Dr.Sumaiya Pathan, Assistant Professor and SoCSE for the constant
guidance during interaction and reviews.
We extend our acknowledgment to the reviewers for critical suggestions and inputs. We also
thank Project coordinator Dr. Uday Kulkarni, and reviewers for their suggestions during the
course of completion. We express gratitude to our beloved parents for constant encouragement
and support.

Sinchan H - 01FE22BCS055
Varda I Pattanshetty- 01FE22BCS189
Varsha I Pattanshetty - 01FE22BCS191
Sujal V Naduvinamani - 01FE22BCS248

i
 CONTENTS

ABSTRACT i

ACKNOWLEDGEMENT i

CONTENTS iii

LIST OF TABLES iv

LIST OF FIGURES v

1 INTRODUCTION 1
1.1 Preamble (Provide Introduction of the project) . . . . . . . . . . . . . . . . . . 1
1.2 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.3 Objectives of the project . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.4 Literature Review / Survey . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.5 Problem Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2

2 SOFTWARE REQUIREMENT SPECIFICATION 3

2.1 Overview of SRS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2.2 Requirement Specifications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2.2.1 Functional Requirements . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2.2.2 Use Case Diagrams . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2.2.3 Use Case Descriptions Using Scenarios (Pressman Template) . . . . . . 4
2.2.4 Nonfunctional Requirements . . . . . . . . . . . . . . . . . . . . . . . . 4
2.3 Software and Hardware Requirement Specifications . . . . . . . . . . . . . . . 4

3 PROPOSED SYSTEM 6
3.1 Description of Proposed System . . . . . . . . . . . . . . . . . . . . . . . . . . 6
3.2 Description of Target Users . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
3.3 Advantages/Applications of Proposed System . . . . . . . . . . . . . . . . . . 7
3.4 Scope (Boundary of Proposed System) . . . . . . . . . . . . . . . . . . . . . . 8

4 SYSTEM DESIGN 10
4.1 Architecture of the System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
4.2 Class Diagram (with Brief Explanation) . . . . . . . . . . . . . . . . . . . . . 10
4.3 Sequence Diagram (with Brief Explanation) . . . . . . . . . . . . . . . . . . . 11

ii
 4.4 ER Diagram and Schema (if Applicable) . . . . . . . . . . . . . . . . . . . . . 11
4.5 State Transition Diagram (if Applicable) . . . . . . . . . . . . . . . . . . . . . 12
4.6 Dataset Description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

5 IMPLEMENTATION 13
5.1 Proposed Methodology (Explain the methodology with diagrams/ Algorithms/flowchart
explanation) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
5.2 Description of Modules (Briefly describe important modules Modules descrip-
tions Module Name : module( ) Input : Output : Describe Code/ Pseudocode/
flowchart . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

6 TESTING 14
6.1 Test Plan and Test Cases Explain in brief the types of testing done. Acceptance
test plan test cases Unit test plan test cases . . . . . . . . . . . . . . . . . . 14

7 RESULTS DISCUSSIONS 15

8 CONCLUSIONS AND FUTURE SCOPE 16

REFERENCES 17

9 Plagiarism Report 18

iii
LIST OF TABLES

iv
 LIST OF FIGURES

3.1 Proposed System Framework . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

3.2 Advantages of Proposed System . . . . . . . . . . . . . . . . . . . . . . . . . . 8
3.3 Scope and Boundary of the Proposed System . . . . . . . . . . . . . . . . . . . 8

4.1 System Architecture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

4.2 Sequence Diagram . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
4.3 State Transition Diagram . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
4.4 Dataset Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

v
 A Deep Learning Model For Post-Disaster Flood Mapping In SAR Images

Chapter 1

INTRODUCTION
Floods cause severe damage and loss of life, making disaster management a challenging
task. Quick and accurate flood mapping is essential for guiding effective response and recovery
efforts. However, traditional flood mapping methods face significant challenges such as cloud
cover, heavy rain, and poor visibility. Synthetic Aperture Radar (SAR) technology overcomes
these limitations by providing high-resolution images irrespective of weather conditions or
time of day. [?]

1.1 Preamble (Provide Introduction of the project)

This project focuses on the development of a deep learning model for post-disaster flood
mapping using SAR images. SAR images are captured using radar sensors that emit microwave
signals and record the reflected signals. These images enable reliable flood detection even
during adverse weather conditions. Deep learning techniques are employed to enhance the
accuracy of flood mapping, providing valuable support for disaster response teams.

1.2 Motivation
Flooding is one of the most devastating natural disasters, causing significant economic losses
and disruption to communities. Accurate and timely flood mapping is critical to mitigate
the impacts and plan effective relief operations. Traditional methods, which rely on optical
images, are often hindered by poor visibility due to clouds or rain. The use of SAR images
and advanced deep learning models offers a promising solution to these challenges, inspiring
the development of this project.

1.3 Objectives of the project

The objectives of this project are as follows:

• To preprocess SAR images for improved model input quality.

• To train a deep learning model capable of identifying flooded areas.

School of Computer Science & Engineering, KLE Technological University, Hubballi - 31 1

 A Deep Learning Model For Post-Disaster Flood Mapping In SAR Images

• To evaluate the model’s accuracy in generating reliable flood maps.

• To ensure the model processes input images within an acceptable time frame.

1.4 Literature Review / Survey

Several studies have explored the use of SAR images and deep learning models for disaster
management. A key advantage of SAR technology is its ability to capture data during adverse
weather conditions, making it a reliable tool for flood mapping. Research has demonstrated
the effectiveness of U-Net and CycleGAN architectures for segmentation and image refinement
tasks. These findings underscore the potential of integrating these models for post-disaster
flood mapping. [?]

1.5 Problem Definition

The problem addressed in this project is to develop an automated deep learning model that
can accurately detect and map flooded areas using SAR imagery. The goal is to provide a
solution that surpasses the limitations of traditional methods, ensuring reliable flood maps
regardless of weather conditions. This model will serve as a critical tool for disaster response
and recovery planning.

School of Computer Science & Engineering, KLE Technological University, Hubballi - 31. 2
 A Deep Learning Model For Post-Disaster Flood Mapping In SAR Images

Chapter 2

SOFTWARE REQUIREMENT
SPECIFICATION
This chapter outlines the software requirements necessary for implementing the project
titled "A Deep Learning Model for Post-Disaster Flood Mapping in SAR Images." The re-
quirements include functional and non-functional aspects, as well as software and hardware
specifications. These are defined to ensure a clear understanding of the project’s goals and
constraints.

2.1 Overview of SRS

The Software Requirement Specification (SRS) provides a detailed description of the sys-
tem’s behavior, features, and constraints. It serves as a guide for developers and stakeholders
to understand the project’s scope and deliverables. The SRS ensures that the system aligns
with the intended functionality and performance requirements.

2.2 Requirement Specifications

The project requirements are categorized into functional and non-functional requirements.
Functional requirements describe the specific tasks the system must perform, while non-
functional requirements define the system’s performance, usability, and other quality at-
tributes.

2.2.1 Functional Requirements

• FR1: The system must accept SAR images as input.

• FR2: The model must detect flooded regions in SAR images with high accuracy.

• FR3: The system should generate a visual map highlighting the flooded areas.

• FR4: The model should provide evaluation metrics for accuracy, precision, and recall.

School of Computer Science & Engineering, KLE Technological University, Hubballi - 31 3

 A Deep Learning Model For Post-Disaster Flood Mapping In SAR Images

2.2.2 Use Case Diagrams

Use case diagrams illustrate the interaction between users and the system for specific func-
tionalities. For this project, the primary use case involves flood detection and mapping.

2.2.3 Use Case Descriptions Using Scenarios (Pressman Template)

The use cases are detailed using scenarios that describe user interactions step-by-step, as per
the Pressman template.

2.2.4 Nonfunctional Requirements

• NFR1: The model must achieve a segmentation accuracy of at least 85%.

• NFR2: The system should process each input image within 5 seconds.

• NFR3: The system should be scalable to handle large datasets efficiently.

• NFR4: The interface should be user-friendly and intuitive for disaster response teams.

Performance Requirements

The system should be capable of processing SAR images in near-real-time to ensure timely
flood mapping.

Security Requirements

The system must ensure data integrity and secure storage of SAR images and flood maps.

Usability

The system should be easy to use, with clear instructions and minimal user training required.

Any Other Requirements

The system must be portable and compatible with various platforms, including cloud and
on-premise setups.

2.3 Software and Hardware Requirement Specifications

The project requires specific software and hardware components to function efficiently.
These requirements ensure that the system meets the performance and scalability demands
of flood mapping.

School of Computer Science & Engineering, KLE Technological University, Hubballi - 31. 4
 A Deep Learning Model For Post-Disaster Flood Mapping In SAR Images

• Software Requirements:

– Python 3.8 or higher with libraries such as TensorFlow, Keras, NumPy, OpenCV,
and Albumentations.
– Jupyter Notebook or Google Colab for model development and testing.
– A GIS tool (optional) for visualizing flood maps.

• Hardware Requirements:

– Minimum 8GB RAM (16GB recommended).

– NVIDIA GPU with CUDA support (e.g., GTX 1060 or higher) for accelerated
model training.
– Storage capacity of at least 100GB for datasets and model files.

School of Computer Science & Engineering, KLE Technological University, Hubballi - 31. 5
 A Deep Learning Model For Post-Disaster Flood Mapping In SAR Images

Chapter 3

PROPOSED SYSTEM
The proposed system aims to automate flood mapping using Synthetic Aperture Radar
(SAR) images and deep learning techniques. It combines the strengths of U-Net and Cycle-
GAN architectures to deliver accurate and efficient flood detection and mapping solutions.

3.1 Description of Proposed System

The system is designed to process SAR images, detect flooded areas, and generate visual
maps of the affected regions. It utilizes U-Net for flood detection due to its efficient segmen-
tation capabilities and CycleGAN to enhance the quality of radar images by reducing noise
and improving clarity.

Figure 3.1: Proposed System Framework

School of Computer Science & Engineering, KLE Technological University, Hubballi - 31 6

 A Deep Learning Model For Post-Disaster Flood Mapping In SAR Images

The system consists of the following modules:

• Data Preprocessing: This module handles resizing, normalization, and augmentation

of SAR images to prepare them for model training.

• Flood Detection with U-Net: U-Net is used for pixel-wise segmentation of SAR
images to identify flooded regions.

• Image Refinement with CycleGAN: CycleGAN enhances SAR images by reducing

noise, enabling better flood detection results.

• Visualization and Mapping: This module generates flood maps by overlaying the
segmented regions on geographic layouts for better interpretability.

The Figure 3.1 illustrates the proposed system’s workflow, highlighting its key components
and their interactions.

3.2 Description of Target Users

The target users of this system include disaster management teams, government authorities,
and researchers involved in flood response and recovery planning. The system is designed
with simplicity and efficiency to ensure usability by non-technical users, while also meeting
the advanced analytical needs of researchers.

3.3 Advantages/Applications of Proposed System

The proposed system offers several advantages over traditional flood mapping methods:

• Weather-Independent Mapping: SAR images enable flood detection irrespective of

cloud cover or lighting conditions.

• High Accuracy: Deep learning models like U-Net ensure precise segmentation of
flooded areas.

• Real-Time Processing: The system processes SAR images quickly, allowing for near-
real-time flood mapping.

• Noise Reduction: CycleGAN enhances image quality, improving the reliability of flood
detection.

School of Computer Science & Engineering, KLE Technological University, Hubballi - 31. 7
 A Deep Learning Model For Post-Disaster Flood Mapping In SAR Images

Figure 3.2: Advantages of Proposed System

The design principles focus on robustness, scalability, and user-friendliness. By leveraging

SAR technology and advanced deep learning methods, the system ensures reliable performance
in diverse scenarios.

3.4 Scope (Boundary of Proposed System)

The scope of the system includes:

• Handling SAR datasets with varying resolutions and formats.

• Segmenting flooded regions with an accuracy of over 85%.

• Generating visual flood maps suitable for disaster response planning.

• Limiting processing time for each image to under 5 seconds.

Figure 3.3: Scope and Boundary of the Proposed System

School of Computer Science & Engineering, KLE Technological University, Hubballi - 31. 8
 A Deep Learning Model For Post-Disaster Flood Mapping In SAR Images

The design principles identified, including modular architecture and robust image prepro-
cessing, ensure the system’s suitability for real-world flood mapping applications. The Figure
3.3 highlights the boundaries and operational scope of the proposed system.

School of Computer Science & Engineering, KLE Technological University, Hubballi - 31. 9
 A Deep Learning Model For Post-Disaster Flood Mapping In SAR Images

Chapter 4

SYSTEM DESIGN
This chapter provides a detailed description of the system’s design and implementation.
Each component is outlined with its corresponding algorithm, diagrams, and dataset descrip-
tions to ensure a comprehensive understanding of the system’s architecture.

4.1 Architecture of the System

The architecture integrates the use of U-Net and CycleGAN models for efficient flood mapping.
The U-Net architecture handles flood region segmentation, while CycleGAN refines SAR image
quality to enhance segmentation accuracy.
Algorithm 1 describes the flood region segmentation process using U-Net.

Algorithm 1 Flood Region Segmentation Using U-Net

Require: SAR images as input
Ensure: Segmented image with flooded regions highlighted
1: Preprocess the SAR images (resize, normalize, augment).
2: Pass the preprocessed images through the U-Net model.
3: Extract feature maps using convolutional layers.
4: Perform upsampling and concatenation for segmentation.
5: Generate the segmented image where:
• White pixels represent water bodies.
• Black pixels represent non-water bodies.
return Segmented image with flooded regions.

images/SystemArchitecture.png

Figure 4.1: System Architecture

4.2 Class Diagram (with Brief Explanation)

The class diagram represents the logical structure of the system, including key entities such
as image preprocessing, model training, and evaluation modules.

School of Computer Science & Engineering, KLE Technological University, Hubballi - 31 10

 A Deep Learning Model For Post-Disaster Flood Mapping In SAR Images

Algorithm 2 demonstrates the implementation of CycleGAN for SAR image refinement.

Algorithm 2 SAR Image Refinement Using CycleGAN

Require: SAR images (unpaired datasets)
Ensure: Enhanced SAR images
1: Define two generators (GAB , GBA ) and two discriminators (DA , DB ).
2: Train the generators to produce refined images:
• GAB : Translates SAR images to enhanced versions.
• GBA : Reconstructs the original SAR images.
3: Train the discriminators to distinguish real and fake images. return Enhanced SAR
images for better segmentation.

4.3 Sequence Diagram (with Brief Explanation)

The sequence diagram outlines the interaction flow between modules. It includes:

• Data preprocessing

• Model inference (U-Net and CycleGAN)

• Visualization and result generation

images/SequenceDiagram.png

Figure 4.2: Sequence Diagram

4.4 ER Diagram and Schema (if Applicable)

The Entity-Relationship (ER) diagram represents the relationships between datasets and the
processing modules. Key entities include:

• SAR Images

• Flood Segmentation Results

• Model Metadata

School of Computer Science & Engineering, KLE Technological University, Hubballi - 31. 11
 A Deep Learning Model For Post-Disaster Flood Mapping In SAR Images

4.5 State Transition Diagram (if Applicable)

The state transition diagram illustrates the system’s state changes, such as:

• Loading data

• Processing images

• Generating results

• Completing evaluation

images/StateTransitionDiagram.png

Figure 4.3: State Transition Diagram

4.6 Dataset Description

The dataset used for this project consists of SAR images obtained from the Sentinel-2 satellite.
The dataset details include:

• Image Size: Images of varying resolutions.

• Training Data: 2000 images (75% of the dataset).

• Testing Data: 683 images (25% of the dataset).

• Preprocessing: Resizing to 256 × 256, normalization to [0, 1], and augmentation for
robustness.

images/DatasetOverview.png

Figure 4.4: Dataset Overview

School of Computer Science & Engineering, KLE Technological University, Hubballi - 31. 12
 A Deep Learning Model For Post-Disaster Flood Mapping In SAR Images

Chapter 5

IMPLEMENTATION
Lorem Ipsum is simply dummy text of the printing and typesetting industry. Lorem Ipsum
has been the industry’s standard dummy text ever since the 1500s, when an unknown printer
took a galley of type and scrambled it to make a type specimen book. It has survived not only
five centuries, but also the leap into electronic typesetting, remaining essentially unchanged.
It was popularised in the 1960s with the release of Letraset sheets containing Lorem Ipsum
passages, and more recently with desktop publishing software like Aldus PageMaker including
versions of Lorem Ipsum. [1]

5.1 Proposed Methodology (Explain the methodology with

diagrams/ Algorithms/flowchart explanation)

5.2 Description of Modules (Briefly describe important

modules Modules descriptions Module Name : mod-
ule( ) Input : Output : Describe Code/ Pseudocode/
flowchart

Algorithm 3 Generating Story Index

Require: Stories represented as s.
Ensure: R min-heap of size k (Rs1 , Rs2 , Rs3 , ..., Rsn ) for all stories.
1: Let L1 , L2 , L3 , ..., L|t| be the posting list of tweets in time t.
2: for all Stories s ∈
S
Li do
3: Attempt inserting (s,CS(s,u)) into Rs
4: end for
5: return (Rs1 , Rs2 , Rs3 , ..., Rsn )

School of Computer Science & Engineering, KLE Technological University, Hubballi - 31 13

 A Deep Learning Model For Post-Disaster Flood Mapping In SAR Images

Chapter 6

TESTING
Lorem Ipsum is simply dummy text of the printing and typesetting industry. Lorem Ipsum
has been the industry’s standard dummy text ever since the 1500s, when an unknown printer
took a galley of type and scrambled it to make a type

6.1 Test Plan and Test Cases Explain in brief the types
of testing done. Acceptance test plan test cases Unit
test plan test cases

School of Computer Science & Engineering, KLE Technological University, Hubballi - 31 14

 A Deep Learning Model For Post-Disaster Flood Mapping In SAR Images

Chapter 7

RESULTS DISCUSSIONS
(Discus the results and its interpretation. Snap shots with brief explanation about the
results Summarize /Aanalyze the results

School of Computer Science & Engineering, KLE Technological University, Hubballi - 31 15

 A Deep Learning Model For Post-Disaster Flood Mapping In SAR Images

Chapter 8

CONCLUSIONS AND FUTURE SCOPE

• Conclude the project by mentioning the features or services you have Implemented. • In
Future work explain how your work can be enhanced • Write conclusion and future work in
different paras.

School of Computer Science & Engineering, KLE Technological University, Hubballi - 31 16

 A Deep Learning Model For Post-Disaster Flood Mapping In SAR Images

REFERENCES
[1] Marcus Fontoura Alexander Shraer, Maxim Gurevich and Vanja Josifovski. “Top-k
Publish-Subscribe for Social Annotation of New". Proceedings of the VLDB Endowment,
pages 6(6):385–396, 26th August 2013.

School of Computer Science & Engineering, KLE Technological University, Hubballi - 31 17

 A Deep Learning Model For Post-Disaster Flood Mapping In SAR Images

Chapter 9

Plagiarism Report
Attach your plagiarism report of this mini report here. Make sure that plagiarism is below
20 %.

School of Computer Science & Engineering, KLE Technological University, Hubballi - 31 18