CHANGE DETECTION IN SENTINEL IMAGES

SENTINEL-1(SAR) SENTINEL-2(RGB)

Shravani Shekar Vaishnavi Kamath Team-Member-3

Team-Member-4
1BM21CS205 1BM21CS235 USN
USN

Ramya KM Dr Umadevi V
Assistant Professor Professor
Department of CSE Department of CSE
B.M.S. College of Engineering B.M.S. College of Engineering

11/12/2024 Title of the Project Department of CSE, BMSCE 1

OUTLINE
1. Introduction
2. Problem Statement
3. Objectives
4. Sustainable Development Goals
5. Collaboration
6. Literature Survey
7. Research Gap Identification
8. High Level Design
9. Technology
10. Datasets
11. Gantt chart for Major Project Phase 1
12. References
11/12/2024 Title of the Project Department of CSE, BMSCE 2
Introduction
Remote sensing change detection involves analysing satellite images taken
over the same geographic area at different times to detect changes such as
land cover transitions, urban development, deforestation or environmental
degradation
The project focuses on:
1.Change Detection: Identifying land use changes through DInSAR processed
Sentinel-1 imagery.
2.Data Integration: Merging DInSAR data with optical Sentinel-2 images for a
holistic view of changes.
3.Deep Learning: Applying deep learning and transfer learning techniques for
automated classification of changes.
11/12/2024 Title of the Project Department of CSE, BMSCE 3
Problem Statement
Accurate detection of surface changes, such as land use transformations and
environmental modifications, is critical for urban planning, disaster management, and
environmental monitoring. However, integrating Sentinel-1 SAR data, processed using
Differential Interferometric Synthetic Aperture Radar (DInSAR) techniques, with
Sentinel-2 optical imagery remains a significant challenge. While Sentinel-1 provides
valuable displacement information, it suffers from coherence loss in vegetated areas
and noise in displacement maps, while Sentinel-2, being multispectral, aids in visual
verification but does not capture displacement. The differences in spatial, spectral, and
temporal characteristics of these datasets complicate their effective fusion, hindering
automated land use change detection. Therefore, there is a need for an integrated
framework that combines DInSAR-processed Sentinel-1 data with Sentinel-2 imagery,
using advanced machine learning and deep learning techniques, to improve the
accuracy and automation of surface change classification.

11/12/2024 Title of the Project Department of CSE, BMSCE 4

Objectives
Acquire and process Sentinel-1B SAR imagery using DInSAR techniques for
terrain-corrected displacement maps.
Verify changes with Sentinel-2 optical imagery to identify industrial
development and surface modifications.
Apply the Laplacian of Gaussian (LoG) algorithm to detect uplift or subsidence
from displacement maps.
Create a balanced dataset combining Sentinel-1 and Sentinel-2 imagery for
machine learning workflows.
Implement advanced machine learning and deep learning models to classify
changes as "changed" or "unchanged."

11/12/2024 Title of the Project Department of CSE, BMSCE 5

Sustainable Development Goals
(SDGs) Addressed
SDG 9: Industry, Innovation, and Infrastructure
It employs advanced technologies including satellite imagery and change detection
algorithms based on machine learning to stimulate innovation in monitoring
industrial change. This supports the development of resilient infrastructure and
fosters innovation in urban planning and construction.

SDG 11: Sustainable Cities and Communities

This project focuses on monitoring urban development and construction activities
through satellite imagery and change detection techniques. By analyzing how cities
evolve, it provides insights into urban planning and sustainable development
practices, ensuring that urban growth is managed in an environmentally friendly and
socially inclusive way.
11/12/2024 Title of the Project Department of CSE, BMSCE 6
Sustainable Development Goals
(SDGs) Addressed
SDG 15:Life on Land
It deals with loss of biodiversity and habitat destruction through land use
changes and their impact on local ecosystems. It suggests that there is a
need for sustainable management of land use by showing an estimation of
the impacts of urbanization and industrial activities.

SDG 12:Responsible Consumption and Production

The project is raising awareness about responsible consumption and
production processes through education of people about the environmental
impact of industrial activity and urbanization.

11/12/2024 Title of the Project Department of CSE, BMSCE 7

Collaboration
• The project should belong to the category of either "Community
Oriented Project" or "Multidisciplinary Project" or Both.

• The project should have field work or collaboration with Hospital,

Police Department, NGO, Agriculture institute, Public sectors like
DRDO, ISRO, BEL, any other academic institute etc,.

11/12/2024 Title of the Project Department of CSE, BMSCE 8

 Literature Survey
Sl. No. Paper Title Year of Problem Statement Methodology Followed Quantitative Future Work Complete
Publication Addressed results Proposed Reference
1. Unsupervised 2020 Detecting and clustering Developed an The best Improving graph Kalinicheva, E.,
Change non-trivial changes in high- unsupervised framework quantitative construction for Ienco, D., Sublime,
Detection resolution satellite image combining autoencoder- results were detecting J. and Trocan, M.,
Analysis in time series (SITS) without based bitemporal achieved for subprocesses 2020.
Satellite Image relying on regular temporal change detection, graph- bitemporal within larger Unsupervised
Time Series resolution or labeled data. based evolution change detection change processes change detection
Using Deep modeling, and clustering on the and enhancing analysis in
Learning using GRU autoencoders Montpellier multitemporal satellite image
Combined and hierarchical dataset, with a filtering of outliers time series using
With Graph- agglomerative precision of 72% directly during deep learning
Based clustering. and a recall of graph combined with
Approaches 79%. construction. graph-based
approaches. IEEE
Journal of
Selected Topics in
Applied Earth
Observations and
Remote Sensing,
13, pp.1450-1466.
2. Change 2022 This study tackles the -Machine learning Validation Further study Goswami, A.,
Detection in problem of change approach through a accuracy up to might involve Sharma, D.,
Remote detection in remote sensing decision tree algorithm implementing Mathuku, H.,
Sensing Image imagery, focusing on the with a separability 100 % with more advance Gangadharan,
Data improvement of matrix for post- the use of machine learning S.M.P., Yadav, C.S.,
Comparing classification accuracy and classification machine models in addition Sahu, S.K.,
Algebraic and efficiency. It compares comparison. learning to better Pradhan, M.K.,
Machine machine learning (decision - Algebraic method of technique and threshold Singh, J. and
Learning tree algorithm) and using image differencing the testing selection Imran, H., 2022.
Methods
11/12/2024 algebraic techniques (image
Title of the Project and theDepartment
corner method
of CSE, BMSCE techniques to Change9 detection
differencing) to detect for threshold selection. accuracy from provide a highly in remote sensing
Sl. No. Paper Title Year of Problem Statement Methodology Followed Quantitative Future Work Complete
Publication Addressed results Proposed Reference
3. Deep Learning 2020 Land cover classification Utilized Fully - Overall Accuracy -Improve Sefrin, O., Riese,
for Land Cover and change detection using Convolutional Networks (OA): 87.0% classification for F.M. and Keller, S.,
Change Sentinel-2 multispectral (FCN) and FCN combined (LSTM with fixed smaller and urban 2020. Deep
Detection data, addressing challenges with Long Short-Term sequence)- classes such as learning for land
like varying water levels Memory (LSTM) Cohen’s Kappa: Buildings. cover change
and class imbalance. networks. Preprocessing 82.0% (LSTM - Enhance detection. Remote
involved class exclusion fixed)- Best class robustness for Sensing, 13(1),
and shoreline masking. accuracy: 100% different temporal p.78.
for Water Body- datasets.
LSTM approaches - Research open-
outperformed set classification
FCN by 3-5 for unknown land
percentage cover types.
points.

4. Spatial Context 2022 Imrove unsupervised Developed a Sibling On the OSCD - Extend SiROC to Kondmann, L.,
Awareness for change detection (CD) in Regression model for dataset, they other remote Toker, A., Saha, S.,
Unsupervised optical satellite images, Optical Change attained state-of- sensing data Schölkopf, B.,
Change addressing challenges like Detection using Half the-art types, such as Leal-Taixé, L. and
Detection in registration errors and Sibling Regression with performance with synthetic aperture Zhu, X.X., 2021.
Optical varying acquisition distn. neighbors to 88.31% specificity, radar. Spatial context
Satellite conditions model pixel values. It is 70.71% sensitivity - Integrate SiROC awareness for
Images enhanced with with deep- unsupervised
ensemble voting from learning-based change detection
mutually exclusive methods for in optical satellite
neighborhoods and enhanced images. IEEE
morphological pseudolabeling Transactions on
operations for object- and self- Geoscience and
level detection. supervised Remote Sensing,
training. 60, pp.1-15.

11/12/2024 Title of the Project Department of CSE, BMSCE 10

Sl. No.
Literature Survey
Paper Title Year of Problem Statement Methodology Followed Quantitative Future Work Complete
Publication Addressed results Proposed Reference
5. DASNet: Dual 2021 Improve the ability of high- DASNet: Dual attentive The results were -Apply DASNet to Chen, J., Yuan, Z.,
Attentive Fully resolution satellite image fully convolutional obtained on the other remote Peng, J., Chen, L.,
Convolutional detection: remove Siamese network. A CDD and BCDD sensing data Huang, H., Zhu, J.,
Siamese pseudochanges and class weighted double-margin datasets. types, like SAR Liu, Y. and Li, H.,
Networks for imbalance contrastive (WDMC) loss CDD Dataset: F1 images. 2020. DASNet:
Change with a dual attention score: 91.9%. - Integrate with Dual attentive
Detection in mechanism (spatial and BCDD Dataset: F1 deep learning fully convolutional
High- channel attention) was score: 89.8% methods for Siamese networks
Resolution used to handle long- pseudo-labeling for change
Satellite range dependencies and self- detection in high-
Images caused by class supervised resolution
imbalance. learning. satellite images.
IEEE Journal of
Selected Topics in
Applied Earth
Observations and
Remote Sensing,
14, pp.1194-1206.
6. Change 2021 Challenges in remote Proposed a Change Achieved superior Reduction of Xu, Q., Chen, K.,
Capsule sensing image change Capsule Network (CCN) F-measure model parameters Zhou, G. and Sun,
Network for detection, including small leveraging capsule (69.1%) and and inference X., 2021. Change
Optical Remote training data sizes, networks for better Kappa (67.7%) on time to improve capsule network
Sensing Image variations in viewpoints, feature extraction and AUAB dataset efficiency and for optical remote
Change and inadequate similarity similarity measurement, compared to practicality for sensing image
Detection measures. designed modules for baseline methods, large-scale and change detection.
unchanged region indicating real-time Remote Sensing,
reconstruction and enhanced feature applications. 13(14), p.2646.
vector-based separability and
comparisons (cosine and reduced noise in
11/12/2024 Title of the Project difference), and tested
Department change detection.
of CSE, BMSCE 11
on AUAB and SZTAKI
Sl. No. Literature Survey
Paper Title Year of
Publication
Problem Statement
Addressed
Methodology Followed Quantitative
results
Future Work
Proposed
Complete
Reference
7. Urban Change 2022 Urban change detection Proposed a best results were Explore improved Fyleris, T.,
Detection from from aerial images with methodology combining obtained by the normalization Kriščiūnas, A.,
Aerial Images issues such as image quality Convolutional Neural model which methods and Gružauskas, V.,
Using differences across time Networks (CNNs) and includes all three extend datasets Čalnerytė, D. and
Convolutional periods and limited training Transfer Learning. The steps of the for enhanced Barauskas, R.,
Neural data. DeepLabv3 model with training process robustness. 2022. Urban
Networks and ResNet50 backbone was which achieved a Investigate change detection
Transfer pretrained on ImageNet mean intersection reducing manual from aerial
Learning and trained using coarse over union (mIoU) efforts in fine- images using
and fine-tuned datasets of 83.1% and pixel tuning dataset convolutional
with normalization accuracy of preparation. neural networks
techniques. 95.2%. and transfer
learning. ISPRS
International
Journal of Geo-
Information,
11(4), p.246.
8. Deep Learning 2021 Challenges in the change Studied two approaches: Best Improvement of Di Pilato, A.,
Approaches to detection using remote (1) semantic segmentation generalization of Taggio, N.,
Earth sensing, such as limited segmentation based on performance with models to varied Pompili, A.,
Observation availability of datasets UNet-like architecture Siam-wbced locations, solution Iacobellis, M., Di
Change labeled, imbalanced nature with EarlyFusion and model: Precision to the overfitting Florio, A.,
Detection of the data, and sensitivity Siamese models; (2) 95.50%, Recall problem in the Passarelli, D. and
to issues of image classification using CNN 82.14% and F1- segmentation Samarelli, S.,
registration. with the EarlyFusion Score 88.30%. The task, 2021. Deep
method. Loss functions classification improvement in learning
were bce and weighted model achieved preprocessing approaches to
bce with Dice loss Recall ~80.16% onboard for real- Earth Observation
wbced. Data and F1-Score time applications. change detection.
augmentation and k-fold ~76.72%. Remote Sensing,
cross-validation were 13(20), p.4083.
11/12/2024 Title of the Project used. Department of CSE, BMSCE 12
Sl. No.
Literature Survey
Paper Title Year of Problem Statement Methodology Followed Quantitative Future Work Complete
Publication Addressed results Proposed Reference
9. HDFNet: 2021 Change detection in optical Proposed a Hierarchical Achieved state-of- Investigate the Zhang, Y., Fu, L.,
Hierarchical aerial images with Dynamic Fusion Network the-art F1 scores: possibility of Li, Y. and Zhang, Y.,
Dynamic multiscale features and (HDFNet) combining 94.77% on multiscale 2021. HDFNet:
Fusion high-resolution imagery, hierarchical fusion, LEBEDEV and features with Hierarchical
Network for addressing false detection dynamic convolution 88.13% on LEVIR- inadequate dynamic fusion
Change and limited completeness modules, and multilevel CD datasets with training data using network for
Detection in issues in traditional supervision to refine high recall and weakly supervised change detection
Optical Aerial methods. change detection. The precision. learning and in optical aerial
Images model applies the adaptability to images. Remote
encoding-decoding small-scale Sensing, 13(8),
architecture with cross- changes and p.1440.
fusion streams. complex
boundaries.

10. Automatic 2021 Automatic annotation of Combines deep learning The U-Net model Extend to larger Neptune, N. and
Annotation of change detection images in for binary change achieved an F1 datasets with Mothe, J., 2021.
Change forested areas using detection using U-Net score of 82% and diverse change Automatic
Detection satellite imagery to provide with ResNet encoder a mean types, use full-text annotation of
Images semantic labels for and visual semantic Intersection over corpora instead of change detection
unlabeled image pairs. embeddings from Union (mIoU) of abstracts, explore images. Sensors,
scientific publications for 70% using datasets from 21(4), p.1110.
automatic labeling. ResNet34 and other sensors
Pretrained and tested on VGG encoders. (e.g., radar), and
Sentinel-2 data of adapt for
Portugal's 2017 unlabeled 3D
wildfires. environmental
datasets.

11/12/2024 Title of the Project Department of CSE, BMSCE 13

Sl. No. Paper Title Year of Problem Statement Methodology Followed Quantitative Future Work Complete
Publication Addressed results Proposed Reference
11. Building 2021 Detection of buildings Combines Pixel-based and Employing U-Net Extend the Jovanović, D.,
Change changed - new, changed, Object-based classifications and OBIA (Object- method to handle Gavrilović, M.,
Detection demolished, using a satellite techniques: U-Net-based Based Image diverse Sladić, D.,
Method to image, to cadastral records Convolutional Neural Network Analysis) methods, geographies and Radulović, A. and
Support and legalization of the and OBIA (eCognition) for the the system building types, Govedarica, M.,
Register of building. segmentation and achieved object improve accuracy 2021. Building
Identified classification. detection accuracy using additional Change Detectio
Changes on between 84% to data sources like Method to
Buildings 88%. DSM and LiDAR, Support Register
and refine of Identified
segmentation Changes on
algorithms to Buildings. Remot
adapt to complex Sensing, 13(16),
urban p.3150.
environments.

12. Satellite On- 2022 Develop an on-board change Proposed an auto-associative AANN - Adapt AANN Guerrisi, G., Del
Board Change detection method for optical neural network (AANN) for outperformed for other Frate, F. and
Detection via satellite images to reduce feature extraction and DWT in change remote Schiavon, G., 202
Auto- data transmission needs and dimensionality reduction. The detection sensing data Satellite on-boar
Associative storage. compressed representations of accuracy. types (e.g., change detection
Neural multi-temporal Sentinel-2 achieved its best SAR). via auto-
Networks images are compared using performance with - Improve associative neura
Euclidean distance (ED) for CVA ground truth automatic networks. Remot
change detection. Compared validation, threshold Sensing, 14(12),
performance with discrete recording an F1 selection for p.2735.
wavelet transform (DWT) score of 0.92 and various land
compression. an AUC of 0.977 cover types.

11/12/2024 Title of the Project Department of CSE, BMSCE 14

Sl. No. Paper Title Year of Problem Statement Methodology Followed Quantitative Future Work Complete
Publication Addressed results Proposed Reference
13. Change 2023 Improve accuracy and Developed Change Tested on four -Extend the Han, C., Wu, C.,
Guiding robustness in change Guiding Network major datasets-, presented CGNet Guo, H., Hu, M.,
Network: detection (CD) for remote (CGNet), enhancing the LEVIR-CD, WHU- to other classes of Li, J. and Chen, H.,
Incorporating sensing imagery, U-Net structure with a CD, SYSU-CD, datasets, such as 2023. Change
Change particularly addressing Change Guide Module S2Looking-CD SAR images guiding network:
Prior to Guide issues with edge integrity (CGM) that uses deep CGNet achieved -Improve on- Incorporating
Change and internal holes in change features to guide the best results on board satellites' change prior to
Detection detected changes. multi-scale feature WHU-CD with an real-time guide change
in Remote fusion. F1 score of 0.925. processing detection in
Sensing performance remote sensing
Imagery imagery. IEEE
Journal of
Selected Topics in
Applied Earth
Observations and
Remote Sensing.
14. EGMT- 2023 Heterogeneous change The model uses a Evaluated on the Extend to real- Xiang, Y., Tian, X.,
CD:Edge- detection between satellite Multimodal Transformer SACD dataset, time data Xu, Y., Guan, X.
Guided and aerial images is with a Spatially Aligned covering acquisition, and Chen, Z.,
Multimodal challenging due to Transformer (SP-T) for Christchurch, New explore synthetic 2023. EGMT-CD:
Transformers differences in resolution image resolution Zealand, the aperture radar Edge-Guided
Change and edge clarity, leading to alignment and an edge model achieved (SAR) to address Multimodal
Detection from inaccurate detection of detection branch to the best results, cloud Transformers
Satellite and changes. improve feature including an F1- interference, and Change Detection
Aerial Images detection, verified on score of 85.72% enhance deep from Satellite and
the SACD dataset. and an IoU of learning models Aerial Images.
75.01%. for better Remote Sensing,
predictive 16(1), p.86.
performance.

11/12/2024 Title of the Project Department of CSE, BMSCE 15

Sl. No. Paper Title Year of Problem Statement Methodology Followed Quantitative Future Work Complete
Publication Addressed results Proposed Reference
15. STCD-EffV2T 2023 the challenge of accurately uses a combination of The model is Exploring Gomroki, M.,
Unet: Semi detecting changes in urban EfficientNetV2 T for evaluated on two alternative data Hasanlou, M. and
Transfer areas using satellite images. feature extraction datasets: one sources like Reinartz, P., 2023.
Learning Existing methods struggle (encoder) and U-Net for from northern synthetic aperture STCD-EffV2T unet:
EfficientNetV2 with feature extraction and creating a change map Iran (Sentinel-2) radar (SAR) semi transfer
T-Unet achieving high accuracy in (decoder). The model is from 2017 to 2021 imagery to learning
Network for complex urban trained on satellite and the Onera overcome EfficientNetV2 T-
Urban/Land environments. images to detect urban Satellite Change challenges in unet network for
Cover Change changes and is Detection dataset optical imagery. urban/land cover
Detection compared with other (OSCD). It Also focusing on change detection
Using models to demonstrate achieved high improving using sentinel-2
Sentinel-2 its effectiveness. overall accuracy of generalization for satellite images.
Satellite 97.66% on the diverse Remote Sensing,
Images Iran dataset. geographies. 15(5), p.1232.
16. HANet: A 2023 Improve accuracy and HANet : This is a Datasets-WHU- - Extend HANet Han, C., Wu, C.,
Hierarchical address data imbalance in hierarchical attention CD, LEVIR-CD. The for other Guo, H., Hu, M.
Attention change detection (CD) using network based on a best quantitative remote and Chen, H.,
Network for high-resolution remote Siamese structure. In results were sensing tasks 2023. HANet: A
Change sensing images this, multiscale features observed on the like SAR image hierarchical
Detection With are integrated, and with WHU-CD dataset, detection. attention network
Bitemporal a Progressive achieving an F1- - Optimize the for change
Very-High- Foreground-Balanced score of 88.49%, PFBS strategy detection with
Resolution Sampling strategy precision of with non- bitemporal very-
Remote (PFBS), imbalances in 88.99%, and recall linear high-resolution
Sensing Images data are also handled. of 88%. approaches. remote sensing
images. IEEE
Journal of
Selected Topics in
Applied Earth
Observations and
Remote Sensing,
16, pp.3867-3878.

11/12/2024 Title of the Project Department of CSE, BMSCE 16

Research Gap identified
Change detection in remote sensing imagery has been a topic of extensive
research, leveraging datasets like Sentinel-1 and Sentinel-2. However, the majority
of studies are concentrated on regions outside India, leaving a significant gap in the
exploration of urban change dynamics within the Indian context. This project
addresses this gap by creating a unique dataset tailored specifically for Indian
urban regions, providing a valuable resource for region-specific analysis.
Moreover, while many existing studies make use of either Sentinel-1 (radar) or
Sentinel-2 (optical) data independently, few have explored their combined use. The
integration of these datasets in this project harnesses their complementary
strengths, offering a more robust framework for detecting changes in complex
urban environments. Unlike most studies that focus on RGB bands of Sentinel-2,
this project utilizes bands 3, 4, and 12, which are particularly suited for urban
change detection due to their spectral characteristics.

11/12/2024 Title of the Project Department of CSE, BMSCE 17

Research Gap identified
Additionally, while deep learning and transfer learning models are common
in the literature, this project introduces the use of Liquid Neural Networks
(LNN), which have not been applied in this domain. LNNs offer unique
advantages in adaptability and efficiency, making them particularly suitable
for handling dynamic and heterogeneous urban landscapes. This innovative
application of LNN further distinguishes the methodology and potential
impact of the study.

By addressing these gaps, this research offers a novel, contextually relevant,

and technically advanced approach to urban change detection in Indian
regions.

11/12/2024 Title of the Project Department of CSE, BMSCE 18

Research Gap identified
 Indian Context: First study to create and analyze a dataset specifically for
Indian urban regions.

 Integration of Sentinel-1 and Sentinel-2: Combining radar and optical data

for enhanced detection.

 Use of Non-RGB Bands: Focus on underutilized Sentinel-2 bands (3, 4, and

12) for urban-specific features.

 Introduction of LNN: First application of Liquid Neural Networks in change

detection, leveraging their adaptability for dynamic urban analysis.

11/12/2024 Title of the Project Department of CSE, BMSCE 19

High Level Design

11/12/2024 Title of the Project Department of CSE, BMSCE 20

Technology to be Used
Software Requirements:
• Remote Sensing Tools
• Sentinel Application Platform (SNAP): For DInSAR processing and generating terrain-
corrected displacement maps.
• GDAL (Geospatial Data Abstraction Library): For geospatial data manipulation and
reprojection.
• Machine Learning and Deep Learning
• TensorFlow and Keras: For building and training deep learning models like Liquid
Neural Networks and EfficientNetB0.
• scikit-learn: For implementing classical machine learning models (e.g., SVM, Random
Forest).
• Data Visualization
• QGIS: For geospatial visualization and overlaying changes on maps.
11/12/2024 Title of the Project Department of CSE, BMSCE 21
Technology to be Used
Hardware Requirements:

1. Processor: Intel i5(12th Gen) 2.5 GHz

2. Installed Ram: 32 GB
3. System Type: x64 bit OS

11/12/2024 Title of the Project Department of CSE, BMSCE 22

Datasets Used
The satellite images used in this project is derived from two satellites Sentinel-
1 and Sentinel-2, part of the European Space Agency's Copernicus program.
Sentinel-1 is a Synthetic Aperture Radar (SAR) satellite. It works in all weather
conditions (day and night), providing C-band SAR data that captures ground
information like surface roughness, moisture, and structural changes.
Sentinel-2 is a multispectral optical satellite. Provides imagery in 13 spectral
bands ranging from visible to short-wave infrared (SWIR), useful for analysing
vegetation, water bodies, and soil characteristics.
URL : https://dataspace.copernicus.eu/

11/12/2024 Title of the Project Department of CSE, BMSCE 23

Gantt chart for Major Project Phase
1

11/12/2024 Title of the Project Department of CSE, BMSCE 24

References
• Karim, Z. and Van Zyl, T.L., 2021. Deep/transfer learning with feature space
ensemble networks (FeatSpaceEnsNets) and average ensemble networks
(AvgEnsNets) for change detection using DInSAR sentinel-1 and optical sentinel-2
satellite data fusion. Remote Sensing, 13(21), p.4394.
• https://asf.alaska.edu/how-to/data-recipes/phase-unwrap-an-interferogram/

11/12/2024 Title of the Project Department of CSE, BMSCE 25

 Thank You

11/12/2024 Title of the Project Department of CSE, BMSCE 26