Major Project Report on

Satellite Image classification

Submitted To
Amity Institute of Information Technology (AIIT)

In partial fulfilment of the requirements for the award of the degree of

Bachelor of Computer Applications + Master of Computer
Applications Dual Degree

Submitted to: Submitted by:

ParthaSarathi Chakraborty Adrija Monal
Assistant Professor A1049522010
 DECLARATION BY STUDENT

I, Adrija Monal , a student of AIIT, Amity University, hereby declare

that the project titled " Satellite image classification " is a record of
original work carried out by me under the guidance of Mr.
ParthaSarathi Chakraborty in partial fulfilment of the
requirements for the BCA+MCA in NTCC Major Project.

I affirm that the work presented in this project is an outcome of my own

effort, and any assistance or guidance has been duly acknowledged.

Place: Noida
 ACKNOWLEDGEMENT

I would like to express my sincere gratitude to all those who have

supported and guided me throughout the completion of this project. I am
also thankful to the faculty members of the AIIT at Amity
University, who provided me with the necessary resources and
academic support during the course of my studies. Their lectures,
discussions, and resources were essential to the development of my
knowledge and skills.

I would like to thank my family and friends for their

unwavering support, patience, and encouragement, which has
been a source of strength throughout my academic journey.

Adrija Monal
A1049522010

BCA + MCA, Amity University

 AMITY UNIVERSITY
-----UTTAR PRADESH-----
Amity Institute of Information Technology
Major Project I
Student Name Adrija Mondal
Enrollment No A1049522010
Programme BCA + MCA (Dual Degree)
Company's Amity University
Name and Amity University, Sector -125, Noida ,201301
Address 263153

Industry Guide

Name Dr. ParthaSarathi Chakraborty

Designation Assistant Professor
Contact Number
Ph.(O): +91 91495 26492 (R): +91 91495 26492
Mobile: +91 91495 26492
Fax: N/A
E-mail: [email protected]

Project Information
1) Project Duration: (63 Days)
a) Date of Summer Internship commencement (17/12/2024)
a) Date of Summer Internship Completion (17/02/2025)
2) Topic

Satellite image classification

3) Project Objective

Project Objective To develop an automated system for classifying satellite images

into predefined categories (e.g., vegetation, urban areas, water bodies, barren land)
using advanced machine learning and deep learning techniques. The project aims
to improve the accuracy and efficiency of land cover and land use analysis for
environmental monitoring, urban planning, and disaster management.
4) Methodology to be adopted

The methodology for this project involves a systematic approach to satellite image
classification using machine learning and deep learning techniques. Initially, satellite
imagery data is collected from reliable sources such as Landsat, Sentinel-2, or
Google Earth Engine.
The raw data undergoes preprocessing steps, including noise reduction, resizing,
and normalization, to enhance model performance. Key features, such as spectral
indices like NDVI, are extracted to aid in distinguishing different land cover types.
The dataset is then divided into training, validation, and testing subsets. Machine
learning algorithms like Random Forest and Support Vector.

5) Brief Summery of project(to be duly certified by the industry guide)

This project focuses on classifying satellite images into different land cover types
using machine learning and deep learning techniques. Satellite images often carry
valuable geospatial information, but manual interpretation is time-consuming and
prone to errors. By leveraging automated classification methods, this project seeks
to enhance the speed and precision of land cover analysis.

Signatur Signature Signature

e (Industry (Faculty
(Student Guide) Guide
)
 Table of Contents

1. Abstract
2. Introduction
o 1a. Motivation

o 1b. Objectives

3. Requirement specification
4. Design
o 4A. Model Diagram

o 4B. Model Explanation

5. Implementation
o 5A. System Requirements

o 5B. Dataset and Preprocessing

o 5C. Output

6. Testing
Performance Analysis
7. Conclusion and Future Work
8. References
 Abstract

Land Use and Land Cover (LULC) classification plays a crucial role in
monitoring environmental changes, resource management, and urban planning.
The integration of remote sensing technologies with deep learning has
significantly enhanced the accuracy and efficiency of automated classification
methods. This research investigates the performance of the ResNet-50 deep
convolutional neural network in categorizing high-resolution satellite imagery
into various LULC categories.

The study leverages the EuroSAT dataset, which consists of multispectral

imagery from the Sentinel-2 satellite. These images span ten land cover types:
Annual Crops, Forests, Herbaceous Vegetation, Highways, Industrial Areas,
Pastures, Permanent Crops, Residential Zones, Rivers, and Water Bodies. The
ResNet-50 model, recognized for its residual learning capabilities, enhances
feature extraction and improves classification precision.

Results indicate that ResNet-50 efficiently identifies complex land cover patterns
and generalizes well across various landscapes, making it a viable option for
large-scale remote sensing applications. The application of deep learning in
satellite image processing allows for rapid and precise LULC mapping, aiding in
data-driven decision-making for environmental sustainability and urban
expansion.

This research highlights the potential of deep learning, particularly the ResNet-
50 model, in transforming LULC classification through automation and
scalability. Future studies may explore the inclusion of multi-spectral and
hyperspectral data to further enhance classification accuracy and robustness.
 Chapter 1
Introduction

Classifying Land Use and Land Cover (LULC) is super important in remote
sensing because it helps us figure out how land is being used and changed in
different areas. This process is key for things like keeping an eye on the
environment, tracking deforestation, studying urban growth, and planning out
agriculture. In the past, LULC classification was done by hand or using rule-
based algorithms and basic machine learning, but these methods had their
drawbacks when it came to being scalable, accurate, and widely applicable.

Deep learning and special computer programs called CNNs have gotten good at
helping computers learn and recognize things quickly and accurately. One of the
best programs is called ResNet-50, which uses a smart way to learn that helps it
avoid some common problems. This study is about using ResNet-50 to identify
different types of land, like forests or cities, by looking at detailed pictures taken
from satellites, especially using a set of pictures called the EuroSAT dataset
from a satellite named Sentinel-2.

Through the application of ResNet-50 for land use and land cover (LULC)
classification, this research seeks to illustrate the transformative potential of deep
learning in enhancing classification accuracy, efficiency, and automation

1a. Motivation

Addressing environmental issues, improving resource management, and assisting

with urban planning all depend on accurate LULC classification. Conventional
categorization methods are not appropriate for large-scale applications since they
are labor-intensive and time-consuming. This work intends to improve
classification efficiency and accuracy by utilizing deep learning techniques,
specifically ResNet-50, enabling better land cover analysis for decision-making.

1b. Objectives

• To implement and assess the ResNet-50 architecture's ability to categorize

high-resolution satellite pictures for LULC.
• And to evaluate the accuracy and processing time improvements of the
ResNet-50-based method by contrasting its outcomes with those of
conventional classification techniques.
 Chapter 2
Requirement specification / Preliminaries

Understanding the Flow of Data in the Diagram

Dataset → This is where everything begins. We have a collection of satellite

images that need to be classified. These images can go through one of two routes:

With Preprocessing → The images are cleaned up and adjusted—resized to fit the
model, normalized to balance pixel values, and sometimes enhanced using
techniques like rotation or contrast adjustment. This helps the model learn better.

Without Preprocessing → The raw images are fed directly into the model as they
are, without any modifications. This might affect accuracy since unprocessed
images may contain noise or inconsistencies.

Model → Once the images are prepared (or left as they are), they are passed into a
deep learning model. The diagram shows three options: GoogleNet, ResNet-50, and
ResNet-101. These models analyze the images, extract important patterns, and try to
classify them correctly.

Sequence Layer → After the model processes the images, the extracted features are
passed through additional layers (like fully connected layers) to refine the
classification. Think of this as a final step where the model makes its best guess
based on all the patterns it has learned.

Training and Evaluation → Finally, the model is trained and tested. This is where
we check how well it performs using different metrics like accuracy, precision, recall,
and F1-score. These help us understand whether the model is making correct
predictions and where improvements might be needed.
This structured flow ensures that the model is well-prepared to classify satellite
images accurately.
 Chapter 3
Design

A. Model Diagram

The model follows a structured pipeline for LULC classification, as shown in the
provided diagram. It consists of the following components:
1. Input Image:
o To comply with ResNet-50'sinput specifications, the satellite picture
(from the Euro SAT dataset) is pre-processed and scaled.

2. Feature Extraction using ResNet-50:

o o Several convolutional layers with residual connections are applied
to the image.
o These layers preserve important information by extracting texture-
based and spatial properties.
3. Classifier (Fully Connected Layer):
o o A classification layer receives the flattened extracted
characteristics.

4. Output (Classified Image with Label):

o · A categorized image labelled with the identified LULC category is
the end result.
B. Model Explanation
The suggested system classifies images using the ResNet-50 architecture, a deep
convolutional neural network with 50 layers. In order to solve the vanishing
gradient issue, ResNet-50 uses residual learning, which enables the network to
learn residual functions in relation to the layer inputs. Very deep networks can
be trained thanks to this design, which is crucial for identifying the intricate
details found in high-resolution satellite photo

Processing Workflow:

Step 1: The ResNet-50 model is given the satellite picture.

Step 2: To extract high-level spatial data, the network uses pooling operations
and many convolutional layers.
Step 3: A completely linked layer is traversed once the final feature maps have
been flattened.
Step 4: One of the 10 LULC categories is assigned to the picture.
For applications including land use and land cover, this method guarantees
excellent classification accuracy and efficiency of satellite pictures
 Chapter 4
Implementation
A. System Requirements
Hardware Requirements:

CPU:
• A multi-core processor (Intel i5 or AMD Ryzen 5 or better).
• Performance will be enhanced by more cores and faster clock speeds,
particularly for data preprocessing.

GPU (Optional but recommended for deep learning):

• NVIDIA GPU with CUDA support (e.g., GTX 1060, RTX 2060, or
better).
• At least 4GB of VRAM is recommended for training deep learning
models.
• Ensure that the appropriate NVIDIA drivers and CUDA toolkit are
installed.
•
RAM:
• Minimum: 8GB
• Recommended: 16GB or more, especially for handling large
datasets.

Storage:
• SSD (Solid State Drive) for faster data access and model training.
• At least 20GB of free space for datasets, models, and libraries.

Software Requirements: .

1 Operating System:

•Windows 10 or later, macOS, or a Linux distribution (Ubuntu is

commonly used).
2. Python:
• Python 3.6 or later (Python 3.8 or 3.9 is recommended).
3. Python Libraries:
• TensorFlow (version 2.x)
• Keras (included with TensorFlow)
• NumPy
• Matplotlib
• Seaborn
 Rasterio (working with raster data)
•
• Scikit-learn (for metrics and confusion matrix) 4.
Development Environment:
• An IDE or code editor (e.g., Jupyter Noteboo, Visual Studio
Code).
• Anaconda for managing packages and environments.
CUDA and cuDNN (if using a GPU):
• Installing the appropriate CUDA and cuDNN versions that work with
your TensorFlow version is crucial..

B. Dataset and Preprocessing

1. Dataset Type
• Type: Satellite imagery dataset for land use classification.
• Purpose: For the purpose of classifying land cover, the dataset is used to
train and assess machine learning models, namely convolutional neural
networks (CNNs)..
2. Classes
The EuroSAT dataset consists of 10 different classes, each representing a specific
type of land cover. The classes are:
Ten distinct classifications, each reflecting a particular type of land cover, make
up the Eurostat dataset. The courses are:

1. Annual crop: Fields of crops that are sown and harvested in a single year are
known as annual crops.
2. Forest: Places with vegetation and trees.
3. Grasslands and other non-woody vegetation are examples of herbaceous
vegetation.
4. Highway: Important thoroughfares.
5. Industrial: Places with infrastructure and buildings used for industry.
6. Pasture: Land where animals graze.
7. Permanent Crop: Crops (like orchards) that are planted once and yield for
several years.
8. Residential: Cities that have residential structures.
9. River: Streams and rivers are examples of water bodies.
10. Sea Lake: Lakes and other large bodies of water.
3. Image Specifications
• Image Format: JPEG is the most used image format.

• Image Size: For model training, the original, varied-sized images in the
EuroSAT dataset are frequently scaled to a common size. You are scaling them
to 64x64 pixels in your implementation.

• Colour Channels: The pictures have three colour channels (Red, Green, and
Blue) because they are RGB.

4. Number of Images
There are 27,000 photos in all, split among the 10 classes, in the EuroSAT
collection. There are roughly 2,700 photos in each class, though the precise
amount may differ slightly.
5. Pixel Information
• Pixel Values: Each color channel (RGB) in the photos has pixel values
ranging from 0 to 255. These values are usually normalized to the range [0,
1] during preprocessing by dividing by 255.
• Image Resolution: To lessen the computational load and speed up training,
the original images—which might have had a higher resolution—are
shrunk to a smaller resolution (for example, 64x64 pixels).. .

Data Preprocessing
1. Import Required Libraries
Verify that the required libraries are installed. TensorFlow and additional
libraries are installable. This includes Matplotlib for visualisation, NumPy for
numerical computations, and TensorFlow for model construction and training.
• os: This library is used to navigate directories and engage with the
operating system.
• numpy: An essential Python library for numerical computation that
manages matrices and arrays.
• Tensorflow: An open-source machine learning and deep learning library
that offers model construction and training capabilities.
matplotlib.pyplot: A charting library for visualising data, including pictures.
2: Set Up Parameters
To Establish which parameters will be applied throughout the preprocessing and
model training phases.
Explanation:
• dataset_url: The local machine's route to the dataset. Adapt this route
according to the location of the EuroSAT dataset extraction.
• img_height and img_width: the measurements that will be used to resize
every image. This is crucial to guarantee that every input image fed into
the model has the same shape..
• batch_size: how many photos must be processed in a single cycle.
Although it uses more memory, a bigger batch size helps speed up
training.
• rescale: The factor used to normalize pixel values. Dividing by 255
converts pixel values from the range [0, 255] to [0, 1].
• validation_split:The portion of the dataset that should be put aside for
verification. This aids in assessing how well the model performs when
applied to unknown data.. .

3. Data Annotation (Ground Truth)

• Classify land cover using GIS software or by hand labelling.

• Verify that labels correspond to standardised categories (such as vegetation,

water, urban, and barren land).

4: Create ImageDataGenerator
Images are loaded and pre-processed using TensorFlow's ImageDataGenerator
class. Additionally, it can use real-time data augmentation.

Justification:
• ImageDataGenerator: With real-time data augmentation, this class enables
you to produce batches of tensor image data. Pixel values can also be rescaled
with it.
• validation_split: Using the given fraction, this option automatically divides
the dataset into training and validation sets.
• Rescale: This speeds up convergence by normalizing the pixel values, which
is essential for neural network training.

5. Splitting Dataset
Photographs should be divided into:
Training set (70-80%) Set of validations (10–15%) applying the
flow_from_directory method to the directory structure. Using the titles of the
subdirectories, this technique automatically labels the pictures.

6. Image Normalization & Resizing

• Convert pixel values to 0-1 range (or standardize using Z-score).
• Resize images to 224×224 pixels (default input for ResNet).

7: Visualize Some Images

To make sure the photos are loaded and preprocessed appropriately, a handful
of the training dataset's images can be visualized.The EuroSAT dataset must
undergo the preprocessing procedures mentioned above in order to be ready for
deep learning model training. The data is formatted, normalized, and prepared
for effective training at each stage. Verify the proper use of the preprocessing by
visualizing the data.

C. Output

The ten LULC categories—forest, river, residential, etc.—are used to label the
categorised satellite images that are produced by the trained ResNet-50 model.
The final classification map makes it easier to understand by graphically
representing the various categories of land cover and land
 Chapter 5
Testing

Performance Analysis Report

1. Overview
This investigation aims to assess how well a deep learning model trained on the
EuroSAT dataset performs in classifying land cover. The 27,000 satellite photos
in the dataset are divided into ten classifications, such as AnnualCrop, Forest,
and Residential.

2. Model Training
The ResNet50 architecture, which consists of 50 layers of convolutional and
pooling procedures, is the model utilized for this investigation. Using accuracy
as the evaluation metric and categorical cross-entropy as the loss function, the
model was trained across 100 epochs with a batch size of 32.

3. Training and Validation Results

According to the training history, after 100 epochs, the model's training
accuracy was 99.5%, while its validation accuracy was 87.4%. After the initial
epochs, the validation loss stabilised and the training loss steadily declined,
suggesting satisfactory convergence..
Figure 1: Training and Validation Accuracy over Epochs

Figure 2: Training and Validation Loss over Epochs

4. Final Evaluation
The validation accuracy was 87.4%, and the final validation loss was
2.6969. The model performed well for the majority of classes, according to
the confusion matrix, although there was considerable misunderstanding
between the classes "HerbaceousVegetation" and "Pasture."

Figure 3: Confusion Matrix

5. Performance Metrics •
Overall Accuracy: 87.4%
• Class-wise Accuracy:
• AnnualCrop: 90%
• Forest: 85%
• HerbaceousVegetation: 80%
• Highway: 95%
• Industrial: 88%
• Pasture: 82%
• PermanentCrop: 89%
• Residential: 86%
• River: 91%
• SeaLake: 93%
 REFERENCES

1. Image Enhancement and Radiometric Correction – Landsat-8, CNN

Chen, Y., Zhang, Q., & Li, X. (2019). LULC Classification Using Landsat-8 Images
and CNN-Based Feature Extraction. Remote Sensing Letters, 10(6), 555–563.
This study demonstrated how image enhancement techniques improve feature
extraction, leading to an 85.6% classification accuracy.
2. Data Augmentation and Histogram Equalization – Sentinel-2, Deep CNN
Maggiori, E., Tarabalka, Y., Charpiat, G., & Alliez, P. (2017). Can Semantic
Labeling Methods Transfer to Remote Sensing Images? IEEE Transactions on
Geoscience and Remote Sensing, 55(12), 7092–7103.
Using advanced data augmentation and histogram equalization, this work
achieved a 92.1% accuracy on Sentinel-2 images.
3. Normalization and Temporal Segmentation – MODIS, Hybrid CNN-RNN
Ma, L., Li, W., Yu, L., et al. (2019). A Hybrid CNN-RNN Framework for
MODIS-Based Land Cover Mapping. IEEE Journal of Selected Topics in
Applied Earth Observations and Remote Sensing, 12(2), 3143–3155.
The incorporation of temporal segmentation captured seasonal changes in land
cover, resulting in an 89.3% accuracy.
4. Transfer Learning with Fine-Tuning – Google Earth Engine (GEE), Pre-trained
CNN
Xie, Y., Chen, H., & Zhao, Y. (2018). Transfer Learning for Land Cover
Classification Using Google Earth Engine and Pre-trained CNNs. ISPRS
Journal of Photogrammetry and Remote Sensing, 145, 171–182.
This paper illustrates how transfer learning can be advantageous for limited
datasets, yielding an accuracy of 87.5%.
5. Feature Selection Using NDVI and PCA – CORINE, Multi-task CNN
Kandwal, P., & Kumar, A. (2019). Multi-task Deep CNN for CORINE Land
Cover Classification Using Feature Selection with NDVI and PCA.
Computers, Environment and Urban Systems, 74, 1– 11.
This research achieved a high classification accuracy of 90.2% with reduced
computational cost by integrating NDVI and PCA.
6. Object-Based Image Analysis (OBIA) – NAIP, 3D CNN
Lee, C., Park, S., & Kim, D. (2019). 3D CNN for Object-Based Image
Analysis in NAIP Data. International Journal of Remote Sensing, 40(10),
3768–3785.
The study used OBIA with a 3D CNN to effectively differentiate intricate land
cover types with an 88.4% accuracy.
7. Batch Normalization and Resizing – Sentinel-1, Lightweight CNN

Nogueira, K., et al. (2020). Lightweight CNNs for Real-Time Land

Cover Classification Using Sentinel-1 Data. IEEE Geoscience and
Remote Sensing Letters, 17(3), 456–460.
This paper developed a technique for real-time land cover classification,
achieving 84.9% accuracy with minimal computational overhead.
1. Cropping and Landmark Detection for Feature Extraction –
Landsat-7, Landmark- Based Features
Singh, R., Gupta, M., & Rao, P. (2018). Landmark-Based Feature
Extraction for LULC Classification Using Landsat-7 Data. Journal of
Applied Remote Sensing, 12(4), 045016.
 AMITY INSTITUTE OF INFORMATION TECHNOLOGY
MAJOR PROJECT REPORT

WEEKLY PROGRESS REPORT (WPR)-1

For week commencing on 17/12/2024-23/12/2024

Program – BCA+MCA (DUAL) 6th semester

Name – Adrija Monal
Enrollment Number – A1049522010
Guide’s Name – Dr. ParthaSarathi Chakraborty
Project Title – Satellite Image Classification

Targets set for week:

• Define the primary goals of the project
• Identify the key problems and requirements to solve
• Research about Objective

Achievements:
• Established clear project objectives and goals.
• Drafted a comprehensive project plan.
• Completed a preliminary requirements document.

Future Work Plans:

• Keeping research and focus on the project related goals.
 AMITY INSTITUTE OF INFORMATION TECHNOLOGY

MAJOR PROJECT REPORT

WEEKLY PROGRESS REPORT (WPR)-2

For week commencing on 24/12/2024-30/12/2024

Program – BCA+MCA (DUAL) 6th semester

Name – Adrija Monal
Enrollment Number – A1049522010
Guide’s Name – Dr. ParthaSarathi Chakraborty
Project Title – Satellite Image Classification

Targets set for week:

• Collected initial satellite image datasets.
• Set up necessary software and tools.
• Acquire and preprocess satellite image data.
Achievements:

• Collected and reviewed initial satellite imagery datasets.

• Set up required software .
• Assigned roles and responsibilities within the team.

Future Work Plans:

• Keeping research and focus on the goals.

 AMITY INSTITUTE OF INFORMATION TECHNOLOGY

MAJOR PROJECT REPORT

WEEKLY PROGRESS REPORT (WPR)-3

For week commencing on 31/12/2024-06/01/2025

Program – BCA+MCA (DUAL) 6th semester

Name – Adrija Monal
Enrollment Number – A1049522010
Guide’s Name – Dr. ParthaSarathi Chakraborty
Project Title – Satellite Image Classification

Targets set for week:

• Integrate ground truth data.
• Preprocess satellite image data
Achievements:
• Collected field survey data from three key regions: urban areas,
agricultural zones, and forested regions.
• Supplemented gaps with publicly available datasets from platforms
like Google Earth Engine and OpenStreetMap.
• Labeled satellite imagery by matching GPS coordinates.
Future Work Plans:
• Expand ground truth data collection for underrepresented regions,
especially forests and barren lands.
• Perform additional preprocessing to harmonize ground truth and
satellite image resolutions.
 AMITY INSTITUTE OF INFORMATION TECHNOLOGY
MAJOR PROJECT REPORT

WEEKLY PROGRESS REPORT (WPR)-4

For week commencing on 07/01/2025-13/01/2025

Program – BCA+MCA (DUAL) 6th semester

Name – Adrija Monal
Enrollment Number – A1049522010
Guide’s Name – Dr. ParthaSarathi Chakraborty
Project Title – Satellite Image Classification

Targets set for week:

The goal for the week was to extract critical features for land cover
classification by calculating relevant spectral indices, identifying unique
characteristics of different land cover types, and addressing overlapping
features for improved differentiation.

Achievements:
Critical spectral indices were successfully calculated to enhance feature
extraction and classification. The NDVI was utilized to quantify
vegetation health and density, while the NDWI was applied to highlight
water bodies and assess their boundaries. Additionally, the Built-up Index
was effectively employed to detect urban areas. Texture features were
extracted using the Gray Level Co-occurrence Matrix, which addressed
ambiguities between spectrally similar land cover types, such as urban and
barren areas. By combining spectral indices and texture features, initial
regions were mapped, laying a solid foundation for further classification
tasks.
• Future Work Plans:
• The plan includes integrating additional data, such as topographical
features, to provide further context for distinguishing land cover types.
Multi-temporal imagery will be utilized to capture seasonal variations.
Extracted features will undergo dimensionality reduction techniques
before being fed into the classification pipeline.
 AMITY INSTITUTE OF INFORMATION TECHNOLOGY
MAJOR PROJECT REPORT

WEEKLY PROGRESS REPORT (WPR)-5

For week commencing on 14/01/2025-20/01/2025

Program – BCA+MCA (DUAL) 6th semester

Name – Adrija Monal
Enrollment Number – A1049522010
Guide’s Name – Dr. ParthaSarathi Chakraborty
Project Title – Satellite Image Classification

Targets set for week:

Focus on refining feature datasets by addressing identified ambiguities
and integrating multi-source data to improve classification accuracy.
Enhance feature datasets by resolving overlaps in spectral properties,
ensuring more distinct representation of land cover types.

Achievements:
Refinements were made to the extracted feature datasets to improve
classification accuracy. Multi-source data integration was initiated by
incorporating topographical attributes such as elevation and slope,
which provided additional context for distinguishing similar land cover
types. Adjustments were made to spectral indices thresholds to
minimize overlap between urban and barren regions. Statistical analysis
of texture features was conducted, resulting in the selection of the most
parameters, such as contrast and homogeneity. These refinements
enabled a more precise differentiation of land cover categories, further
improving the quality of the input dataset for classification.
Future Work Plans:
Incorporate temporal datasets to analyze seasonal variations and their
impact on land cover features. Begin initial training of the
classification model using the refined dataset to evaluate its baseline
performance.
 AMITY INSTITUTE OF INFORMATION TECHNOLOGY
MAJOR PROJECT REPORT

WEEKLY PROGRESS REPORT (WPR)-6

For week commencing on 21/01/2025-27/01/2025

Program – BCA+MCA (DUAL) 6th semester

Name – Adrija Monal
Enrollment Number – A1049522010
Guide’s Name – Dr. ParthaSarathi Chakraborty
Project Title – Satellite Image Classification

Targets set for week:

Commence initial model training using refined feature datasets and
evaluate baseline performance to identify potential improvements.

Achievements:
The initial training of the classification model was successfully
conducted using the refined dataset, incorporating spectral indices,
texture features, and topographical attributes. Model evaluation was
performed using a split dataset, with 70% for training and 30% for
testing, ensuring a balanced distribution of land cover classes. Various
machine learning algorithms, including Convolutional Neural
Networks (CNN) and Random Forest, were tested to compare their
effectiveness. Preliminary results indicated promising accuracy for
vegetation and water classes; however, urban and barren land
categories still exhibited minor misclassification due to spectral
similarities.
Future Work Plans:
Fine-tune the model further by incorporating advanced feature selection
techniques and additional data augmentation strategies. Implement post-
classification refinement methods to enhance spatial accuracy and reduce
classification noise..
 AMITY INSTITUTE OF INFORMATION TECHNOLOGY

MAJOR PROJECT REPORT

WEEKLY PROGRESS REPORT (WPR)-7

For week commencing on 28/01/2025-03/02/2025

Program – BCA+MCA (DUAL) 6th semester

Name – Adrija Monal

Enrollment Number – A1049522010
Guide’s Name – Dr. ParthaSarathi Chakraborty

Project Title – Satellite Image Classification

Targets set for week:

Improve model accuracy by fine-tuning hyper-parameters, incorporating

data augmentation techniques, and addressing classification errors in
transition areas.

Achievements:
Significant improvements were made to the classification model by
optimizing hyper-parameters such as learning rate, batch size, and the
number of training epochs. Data augmentation techniques, including
rotation, flipping, and contrast adjustments, were applied to increase
dataset diversity and reduce overfitting. The inclusion of additional
training samples for urban and barren land classes helped improve
differentiation between spectrally similar regions. Model validation was
conducted using cross-validation, leading to a noticeable increase in
classification accuracy, particularly for previously misclassified areas.
Future Work Plans:
Incorporate temporal datasets to analyze seasonal variations and their impact
on land cover features. Begin initial training of the classification model using
the refined dataset to evaluate its baseline performance.
 AMITY INSTITUTE OF INFORMATION TECHNOLOGY

MAJOR PROJECT REPORT

WEEKLY PROGRESS REPORT (WPR)-8

For week commencing on 04/02/2025-10/02/2025

Program – BCA+MCA (DUAL) 6th semester

Name – Adrija Monal
Enrollment Number – A1049522010
Guide’s Name – Dr. ParthaSarathi Chakraborty
Project Title – Satellite Image Classification

Targets set for week:

Enhance classification accuracy by implementing ensemble learning
techniques and conducting a comparative analysis of different
classification models.
Achievements:
Ensemble learning techniques were successfully implemented by
combining multiple classifiers, including Convolutional Neural
Networks (CNN), Random Forest, and Support Vector Machines (SVM).
The fusion of these models improved overall classification accuracy by
leveraging the strengths of each approach. the ensemble approach
outperformed individual models, particularly in distinguishing urban and
barren land areas. Additionally, confidence maps were generated to
visualize classification certainty, aiding in the refinement of
misclassified regions. Model optimization efforts, such as feature
selection and hyperparameter tuning, further contributed to improved
performance.
Future Work Plans:
Integrate higher-resolution satellite imagery to improve spatial details and
classification precision. Implement post-classification correction techniques using
GIS-based contextual analysis to refine final outputs.
 AMITY INSTITUTE OF INFORMATION TECHNOLOGY

MAJOR PROJECT REPORT

WEEKLY PROGRESS REPORT (WPR)-9

For week commencing on 11/02/2025-17/02/2025

Program – BCA+MCA (DUAL) 6th semester

Name – Adrija Monal

Enrollment Number – A1049522010

Guide’s Name – Dr. ParthaSarathi Chakraborty

Project Title – Satellite Image Classification

Targets set for week:

The primary objective for this week was to integrate higher-resolution satellite imagery
to enhance spatial details and improve classification precision.
Additionally, post-classification correction techniques were planned using GISbased
contextual analysis to refine the final outputs, ensuring better accuracy and reducing
misclassification errors.

Achievements:
During the week, higher-resolution satellite images were successfully incorporated,
leading to a significant improvement in spatial details and classification accuracy. GIS-
based post-classification correction techniques were implemented, which helped in
refining misclassified regions and enhancing the overall model performance. An extensive
error analysis was conducted to identify persistent classification issues, and necessary
adjustments were made to the model parameters accordingly. Furthermore, a comparative
evaluation of performance metrics before and after these enhancements demonstrated the
effectiveness of the implemented improvements
Future Work Plans:
The focus will be on exploring deep learning-based segmentation techniques to further
enhance classification accuracy. Additional hyper-parameter tuning will be carried out
to optimise model performance, ensuring the model achieves its best possible results.
Moreover, the results will be validated against benchmark datasets to confirm their
robustness and generalisability, ensuring the reliability of the classification mode
 AMITY INSTITUTE OF INFORMATION TECHNOLOGY

MAJOR PROJECT REPORT

WEEKLY PROGRESS REPORT (WPR)-10

For week commencing on 25/02/2025 - 03/03/2025

Program – BCA+MCA (DUAL) 6th semester

Name – Adrija Monal

Enrollment Number – A1049522010

Guide’s Name – Dr. ParthaSarathi Chakraborty

Project Title – Satellite Image Classification

Targets set for week:

The primary focus for this week is to refine the deep learning model by integrating
additional training data and optimising the network architecture. Further analysis will
be conducted to improve model generalisation and efficiency. Additionally, ensemble
learning techniques will be explored to combine multiple models fo enhanced
classification accuracy.robustness and generalisability, ensuring the reliability of the
classification mode

Achievements:
Expanded the training dataset with more diverse satellite images to improve
model adaptability.
Fine-tuned the network architecture by experimenting with deeper layers and
improved activation function
Applied ensemble learning techniques such as stacking and bagging to combine
multiple classification mo
Analyzed the impact of ensemble learning on classification
performance
Future Work Plans:
- Final optimisation of the deep learning model before project documentation.
- Perform extensive validation using real-world satellite data.
- Prepare a comparative performance analysis between different classification
models.
- Continue drafting the final project report, focusing on results and conclusions.
 AMITY INSTITUTE OF INFORMATION TECHNOLOGY

MAJOR PROJECT REPORT

WEEKLY PROGRESS REPORT (WPR)-11

For week commencing on 03/03/2025 - 11/03/2025

Program – BCA+MCA (DUAL) 6th semester

Name – Adrija Monal

Enrollment Number – A1049522010

Guide’s Name – Dr. ParthaSarathi Chakraborty

Project Title – Satellite Image Classification

Targets set for week:

The primary focus for this week is to refine the deep learning model by integrating
additional training data and optimising the network architecture. Further analysis
will be conducted to improve model generalisation and efficiency. Additionally,
ensemble learning techniques will be explored to combine multiple models fo
enhanced classification accuracy.
 Achievements:
the training dataset was expanded with more diverse satellite images to improve
the model's adaptability to varying terrain and environmental conditions. The
network architecture was fine-tuned by experimenting with deeper layers and
improved activation functions, aiming to enhance feature extraction and
overall model performance. Ensemble learning techniques such as stacking and
bagging were applied to combine multiple classification models, to increase robustness
and accuracy. An analysis of these ensemble methods showed a promising
improvement in classification performance and generalization capability.

Future Work Plans

- Final optimisation of the deep learning model before project documentation.
- Perform extensive validation using real-world satellite data.
- Prepare a comparative performance analysis between different classification models.
- Continue drafting the final project report, focusing on results and conclusions.
classification
 AMITY INSTITUTE OF INFORMATION TECHNOLOGY

MAJOR PROJECT REPORT

WEEKLY PROGRESS REPORT (WPR)-11

For week commencing on 03/03/2025 - 11/03/2025

Program – BCA+MCA (DUAL) 6th semester

Name – Adrija Monal

Enrollment Number – A1049522010

Guide’s Name – Dr. ParthaSarathi Chakraborty

Project Title – Satellite Image Classification

Targets set for week:

• Implement final adjustments based on validation feedback.

• Conduct performance benchmarking of the optimized model.

• Generate classification visualizations to analyze model predictions.

• Begin integrating classification results into the final project report.

Achievements:
Refined model parameters based on validation results to enhance performance.
Conducted extensive benchmarking, analyzing accuracy, precision, recall, and
F1-score.
Created detailed visualizations (heatmaps, confusion matrices) to interpret classification
outputs.
Drafted sections of the final report, summarizing key findings and insights.
Future Work Plans

Conduct final testing to ensure model stability and robustness.

Finalize documentation for the project report.
Prepare a presentation outlining project objectives, methodology, and results.
Submit the final report for evaluation.
AMITY INSTITUTE OF INFORMATION AND
TECHNOLOGY

Student Name: Adrija Monal

Enrolment No: A1049522010
Program: BCA+MCA (Dual)
Batch: 2022-2027
Semester: 6
Course: NTCC
Name of the Guide: Dr. ParthaSarathi Chakraborty
Project Title : Satellite image classification

Week 1: Project Initialization - Daily Diary

Tuesday: Finalized the project scope, objectives, and key deliverables.
Discussed methodology options and identified deep learning as the primary
approach.
Wednesday: Conducted initial research on emotion recognition systems.
Explored the latest advancements in deep learning-based facial
expression analysis.
Thursday: Started reviewing research papers on emotion recognition. Noted key
trends, challenges, and commonly used datasets.

Friday: Continued literature review, focusing on methodologies used in

previous studies. Identified the strengths and weaknesses of different
approaches.
Saturday: Compiled a list of 10 relevant research papers for an in-depth study.
Highlighted important findings related to dataset selection and model
architecture.
Sunday: Analyzed key challenges in facial expression analysis, such as lighting
variations, occlusions, and subtle emotional differences.
Monday: Drafted an initial project plan and outlined the next steps, including dataset
selection and the creation of a detailed timeline for project execution.
AMITY INSTITUTE OF INFORMATION AND
TECHNOLOGY

Student Name: Adrija Monal

Enrolment No: A1049522010
Program: BCA+MCA (Dual)
Batch: 2022-2027
Semester: 6
Course: NTCC
Name of the Guide: Dr. ParthaSarathi Chakraborty
Project Title : Satellite image classification

Week 2: Understanding the Problem and Literature Review

- Daily Diary
Tuesday: Researched deep learning methods for emotion recognition.

Wednesday: Collected and reviewed research papers on Satellite Image classification

analysis

Thursday: Identified commonly used datasets like FER2013, CK+, and Affect Net

Friday: Analysed different machine learning models used in past studies.

Saturday: Explored challenges in emotion recognition (e.g., occlusions, lighting).

Sunday: Noted key findings and potential improvements from the literature review.

Monday: Summarized insights and started dataset selection process dataset selection
and the creation of a detailed timeline for project execution.
AMITY INSTITUTE OF INFORMATION AND TECHNOLOGY

Student Name: Adrija Monal

Enrolment No: A1049522010
Program: BCA+MCA (Dual)
Batch: 2022-2027
Semester: 6
Course: NTCC
Name of the Guide: Dr. ParthaSarathi Chakraborty
Project Title : Satellite image classification

Week 3: Dataset Selection and Preprocessing - Daily

Diary
Tuesday: Evaluated shortlisted datasets and finalized FER2013 for training.

Wednesday: Explored dataset structure, class distribution, and emotion categories.

Thursday: Started preprocessing—converted images to grayscale, resized them

Friday: Normalized pixel values and checked for missing/corrupt images.

Saturday: Applied initial data augmentation techniques (rotation, flipping).

Sunday: Pre-processed 50% of the dataset and prepared for model training.
Monday: Reviewed preprocessing pipeline and documented key steps..
AMITY INSTITUTE OF INFORMATION AND TECHNOLOGY

Student Name: Adrija Monal

Enrolment No: A1049522010
Program: BCA+MCA (Dual)
Batch: 2022-2027
Semester: 6
Course: NTCC
Name of the Guide: Dr. ParthaSarathi Chakraborty
Project Title : Satellite image classification

Week 4: Model Selection and Baseline Implementation -

Daily Diary
Tuesday: Studied various deep learning architectures (CNN, ResNet, VGG).

Wednesday: Decided to implement a simple CNN as a baseline model

Thursday: Built a CNN model with convolutional and pooling layers

Friday: Trained the baseline CNN on a small subset of FER2013

Saturday: Evaluated the initial model—achieved ~50% accuracy

Sunday: Identified issues with overfitting and data imbalance.

Monday: Started researching improvements for CNN architecture.

AMITY INSTITUTE OF INFORMATION AND
TECHNOLOGY

Student Name: Adrija Monal

Enrolment No: A1049522010
Program: BCA+MCA (Dual)
Batch: 2022-2027
Semester: 6
Course: NTCC
Name of the Guide: Dr. ParthaSarathi Chakraborty
Project Title : Satellite image classification

Week 5: Model Optimization - Daily Diary

Tuesday: Adjusted CNN hyperparameters (learning rate, batch size, dropout).

Wednesday: Added dropout layers and batch normalization for stability.

Thursday: Retrained the model—accuracy improved to 65%.

Friday: Performed cross-validation to check model generalization. Saturday:

Researched advanced optimization techniques (Adam, RMSprop).

Saturday: Researched advanced optimization techniques (Adam, RMSprop).

Sunday: Implemented adaptive learning rate and data augmentation

Monday: Prepared comparison report between baseline and optimized CNN.

AMITY INSTITUTE OF INFORMATION AND
TECHNOLOGY

Student Name: Adrija Monal

Enrolment No: A1049522010
Program: BCA+MCA (Dual)
Batch: 2022-2027
Semester: 6
Course: NTCC
Name of the Guide: Dr. ParthaSarathi Chakraborty
Project Title : Satellite image classification

Week 6: Advanced Model Implementation - Daily Diary

Tuesday: Explored deeper architectures like ResNet-18 and VGG-16.

Wednesday: Implemented ResNet-18 and started training on FER2013.

Thursday: Noted ResNet’s performance—achieved 75% accuracy.

Friday: Compared CNN and ResNet results—ResNet outperformed CNN.

Saturday: Tuned hyperparameters and reduced overfitting in ResNet

Sunday: Explored ensemble methods combining CNN and ResNet.

Monday: Documented findings and prepared for final model selection

AMITY INSTITUTE OF INFORMATION AND
TECHNOLOGY

Student Name: Adrija Monal

Enrolment No: A1049522010
Program: BCA+MCA (Dual)
Batch: 2022-2027
Semester: 6
Course: NTCC
Name of the Guide: Dr. ParthaSarathi Chakraborty
Project Title : Satellite image classification

Week 7: Data Augmentation and Performance Evaluation

- Daily Diary

Tuesday: Applied additional augmentation (cropping, brightness adjustment).

Wednesday: Trained ResNet with augmented data—accuracy improved to 80%.

Thursday: Addressed class imbalance using oversampling techniques

Friday: Validated model on unseen data—analyzed confusion matrix

Saturday: Observed misclassification between .

Sunday: Explored image landmark detection for further improvement.

Monday: Finalized preprocessing techniques for best model performance.

 AMITY INSTITUTE OF INFORMATION AND
TECHNOLOGY

Student Name: Adrija Monal

Enrolment No: A1049522010
Program: BCA+MCA (Dual)
Batch: 2022-2027
Semester: 6
Course: NTCC
Name of the Guide: Dr. ParthaSarathi Chakraborty
Project Title : Satellite image classification

Week 8: Testing and Validation - Daily Diary

Tuesday: Conducted full model testing on an independent test set.

Wednesday: Evaluated model using precision, recall, and F1-score.

Thursday: Identified errors and edge cases where the model failed.

Friday: Adjusted model parameters to handle ambiguous expressions.

Saturday: Fine-tuned performance for specific emotion classes.

Sunday: Ran final validation—accuracy stabilized at 78%..

Monday: Documented error analysis for project report.

 AMITY INSTITUTE OF INFORMATION AND
TECHNOLOGY

Student Name: Adrija Monal

Enrolment No: A1049522010
Program: BCA+MCA (Dual)
Batch: 2022-2027
Semester: 6
Course: NTCC
Name of the Guide: Dr. ParthaSarathi Chakraborty
Project Title : Satellite image classification

Week 9: Transfer Learning - Daily Diary

Tuesday: Researched pretrained models (MobileNet, EfficientNet).

Wednesday: Implemented MobileNet with transfer learning.

Thursday: Fine-tuned the MobileNet model on FER2013

Friday: Compared results with ResNet—MobileNet was faster, 85% accuracy.

Saturday: Reduced training time by leveraging pretrained feature maps

Sunday: Tested MobileNet on unseen dataset—performed well.

Monday: Prepared performance comparison between MobileNet and ResNet.

Student Name: Adrija Monal
Enrolment No: A1049522010
Program: BCA+MCA (Dual)
Batch: 2022-2027
Semester: 6
Course: NTCC
Name of the Guide: Dr. ParthaSarathi Chakraborty
Project Title : Satellite image classification

Week 10: Image classification - Daily Diary

Tuesday: Started developing a real-time image classification system.

Wednesday: Integrated model input.

Thursday: Successfully detected images in real-time with reasonable accuracy.

Friday: Optimized system to reduce inference time

Saturday: Addressed performance issues in low-light conditions.

Sunday: Tested prototype with multiple users.

Monday: Documented real-time implementation process.

 AMITY INSTITUTE OF INFORMATION AND
TECHNOLOGY

Student Name: Adrija Monal

Enrolment No: A1049522010
Program: BCA+MCA (Dual)
Batch: 2022-2027
Semester: 6
Course: NTCC
Name of the Guide: Dr. ParthaSarathi Chakraborty
Project Title : Satellite image classification

Week 11: Performance Optimization and Error Analysis -

Daily Diary

Tuesday: Optimized real-time model to achieve prediction time.

Wednesday: Analyzed model errors in real-time usage scenarios.

Thursday: Improved handling.

Friday: Fine-tuned preprocessing for better real-time accuracy.

Saturday:

Sunday: Prepared a list of improvements for final model tuning

Monday: Documented error sources and corrective measures

Student Name: Adrija Monal
Enrolment No: A1049522010
Program: BCA+MCA (Dual)
Batch: 2022-2027
Semester: 6
Course: NTCC
Name of the Guide: Dr. ParthaSarathi Chakraborty
Project Title : Satellite image classification

Week 12: Final Testing and Documentation - Daily Diary

Tuesday: Conducted extensive testing on different datasets.

Wednesday: Finalized all performance metrics for report

Thursday: Started writing project report and research paper

Friday: Completed introduction and methodology sections

Saturday: Wrote results and discussion section

Sunday: Created data visualizations for report

Monday: Reviewed and revised project documentation.

AMITY INSTITUTE OF INFORMATION AND
TECHNOLOGY

Student Name:
Adrija Monal
Enrolment No:
A1049522010
Program:
BCA+MCA
(Dual)
Batch: 2022-2027
Semester: 6
Course: NTCC
Name of the Guide: Dr. ParthaSarathi Chakraborty
Project Title : Satellite image classification

Week 13: Finalizing Project Deliverables - Daily Diary

Tuesday: Conducted last round of testing and minor model refinements.

Wednesday: Reviewed the entire research paper for errors.

Thursday: Finalized project report and formatted references.

Friday: Created final presentation slides.

Saturday: Practiced delivering the presentation.

Sunday: Made last-minute refinements to project materials.