CERTIFICATE

1
 DECLARATION
I hereby declare that the Industrial Training Report entitled "AI-Powered Deforestation De-
tection Using Satellite Images” is an authentic record of my own work as requirements of
Industry Training during the period from 13th January 2025 to 31st May 2025 for the award
of the degree of B.E. Under the guidance of Dr. Praveen M Dhulavvagol.

Nikita S P
01FE22BCS404

2
 ACKNOWLEDGMENT
The satisfaction and euphoria that accompany the successful completion of any task would be
incomplete without the mention of a number of individuals whose professional guidance and
encouragement helped me in the successful completion of this report work.
I also take this opportunity to thank Dr. Vijaylakshmi M, Professor and Head, School
of Computer Science and Engineering for having provided us academic environment which
nurtured our practical skills contributing to the success of our project.
I sincerely thank our guide Dr. Praveen M Dhulavvagol, School of Computer Science and
Engineering for his guidance and wholehearted co-operation during the course of completion.
I sincerely thank Deepak Ghatteppanavar, HAEGL Technologies Pvt. Ltd. for his support,
inspiration and wholehearted co-operation during the course of completion.
My gratitude will not be complete without thanking our beloved parents, our seniors and
our friends who have been a constant source of aspirations.

Nikita S P
01FE22BCS404

3
 ABOUT COMPANY

Haegl Technologies Pvt. Ltd. is an innovative and fast-growing technology Learning (ML),
and cloud-based systems. Its mission is to simplify and automate enterprise operations across
various sectors including healthcare, education, and agriculture through smart, scalable, and
user-friendly platforms.
One of the company’s flagship offerings is Cloud 360°, an integrated cloud-based platform that
company headquartered in Dharwad, Karnataka, India. Established in 2019, Haegl focuses on
delivering intelligent digital solutions powered by Artificial Intelligence (AI), Machine com-
bines Enterprise Resource Planning (ERP), Customer Relationship Management (CRM), and
Human Resource Management (HRM) modules. Cloud 360° is designed to enhance produc-
tivity, streamline workflows, and enable real-time data insights using AI-driven analytics.
A core part of Haegl’s technology stack includes Django, a high-level Python web framework
known for its security, scalability, and rapid development capabilities. Projects developed us-
ing Django benefit from a robust structure, reusable components, and built-in administrative
tools, making it ideal for building AI-enhanced enterprise applications. By leveraging Django
with AI models, Haegl is able to deliver intelligent features such as predictive analytics, au-
tomated decision-making, and real-time recommendations.
Haegl also introduced the KIAS Framework—which stands for Knowledge, Insight, Automa-
tion, and Systems—to help businesses transition into the digital era with intelligent automa-
tion and advanced system integration.
With a strong focus on innovation, user-centric design, and secure deployment, Haegl Tech-
nologies is emerging as a leading provider of AI-powered enterprise solutions in India.

4
 ABSTRACT
Deforestation poses a significant threat to global ecosystems, contributing to biodiversity
loss, climate change, and environmental degradation. Traditional approaches to monitoring
deforestation, such as manual inspection of satellite imagery, are labor-intensive, slow, and
prone to human error. To address these challenges, this project presents an AI-powered web
application designed to automatically detect deforestation from satellite images.
The system is developed using the Django web framework and incorporates machine learn-
ing techniques for image classification. Users can upload satellite images through a simple web
interface. Once uploaded, the image is processed using a trained classification model that de-
termines whether the area depicted is affected by deforestation. The result is immediately
displayed to the user, providing quick insights into forest conditions.
This application aims to assist environmental researchers, government agencies, and NGOs
in monitoring forest changes in real-time, reducing response times to illegal logging activities,
and enabling data-driven decision-making. With features like automated image analysis, me-
dia storage, and scalable architecture, the system serves as a practical and efficient tool for
environmental surveillance.
The project demonstrates how integrating artificial intelligence with web technologies can
create powerful solutions for environmental challenges, laying the foundation for future en-
hancements such as real-time alerts, geospatial mapping, and deep learning-based segmenta-
tion.

Keywords: Deforestation Detection, Satellite Imagery, Machine Learning, Django Web

Application, AI-Based Classification, Forest Surveillance, Real-Time Analysis, Smart
Dashboard, Geographic Information Systems (GIS)

i
 CONTENTS

Declaration 2

Acknowledgement 3

About Company 4

ABSTRACT i

CONTENTS iii

LIST OF FIGURES iv

1 INTRODUCTION 1
1.1 Literature Survey . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.3 Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.4 Problem Defination . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

2 REQUIREMENT ANALYSIS 6
2.1 System Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.2 Functional Requirements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.3 Non-Functional Requirements . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.4 Database Requirements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

3 SYSTEM DESIGN 10
3.1 Architecture Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
3.2 Data Flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
3.3 Use Case Diagram . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

4 Implementation 16
4.1 Tools and technologies used . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
4.2 Dataset Acquisition and Handling . . . . . . . . . . . . . . . . . . . . . . . . . 17
4.3 Model Development and Web Integration . . . . . . . . . . . . . . . . . . . . . 18

5 Testing 19
5.1 Integration Testing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
5.2 Acceptance Testing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20

ii
6 Results and Discussions 22

7 Conclusion and Future Scope 29

7.1 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
7.2 Future Scope . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

REFERENCES 31

iii
 LIST OF FIGURES

2.1 System Architecture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

3.1 Architecture Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

3.2 Data Flow Diagram . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
3.3 Use Case Diagram . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

4.1 Comparison of categories with tools and technologies . . . . . . . . . . . . . . 17

4.2 Framework of Python Web Development with Django . . . . . . . . . . . . . . 18

5.1 Test Scenario of Integration Testing . . . . . . . . . . . . . . . . . . . . . . . . 20

5.2 Test Scenario of Acceptance Testing . . . . . . . . . . . . . . . . . . . . . . . . 21

6.1 Registration Page . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

6.2 Login Page . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
6.3 Research Dashboard - Uploaded Images & Data . . . . . . . . . . . . . . . . . 23
6.4 Admin Panel - Dashboard Overview . . . . . . . . . . . . . . . . . . . . . . . . 24
6.5 Admin Panel - Researcher Management . . . . . . . . . . . . . . . . . . . . . . 24
6.6 Admin Panel - Analyst Management . . . . . . . . . . . . . . . . . . . . . . . 25
6.7 Analyst Dashboard - Deforestation Trends . . . . . . . . . . . . . . . . . . . . 25
6.8 Upload and Process Page . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
6.9 Dashboard with PDF Downloads . . . . . . . . . . . . . . . . . . . . . . . . . 26
6.10 Analyst Dashboard - Deforestation Trends (Tabular View) . . . . . . . . . . . 27
6.11 Analyst Dashboard - Forecasting Chart . . . . . . . . . . . . . . . . . . . . . . 27
6.12 Admin Panel - Report Section . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

iv
 AI-Powered Deforestation Detection Using Satellite Images

Chapter 1

INTRODUCTION
Deforestation is a pressing global concern that threatens biodiversity, accelerates climate
change, and disrupts ecosystems. With forests being cleared at unprecedented rates for agricul-
ture, infrastructure, and industrial expansion, there is a growing need for advanced technolog-
ical solutions to monitor and mitigate this damage. Traditional methods of forest monitoring
often rely on manual surveys or outdated datasets, which are slow, labor-intensive, and prone
to inaccuracies.
This project aims to address these challenges by developing an AI-powered deforestation
detection system that utilizes satellite imagery and machine learning algorithms to automat-
ically identify and analyze deforested regions. The system is designed to support multiple
user roles, including administrators and analysts, with role-based access control and secure
login functionality. Users can upload satellite images and corresponding masks, which are
then processed by a trained AI model to detect deforestation and highlight affected areas.
In addition to detection, the system offers advanced features such as environmental impact
estimation—calculating the ecological consequences of deforested areas—and a predictive ana-
lytics module to forecast potential future deforestation hotspots based on historical trends. A
comprehensive reporting module is also integrated, enabling the generation of downloadable
reports that include visuals, metrics, and insights. Furthermore, an interactive dashboard
displays analytical results through graphs, charts, and heatmaps for easy interpretation.
This project not only demonstrates the application of AI and geospatial analysis in envi-
ronmental monitoring but also contributes toward building intelligent systems that support
sustainable development and forest conservation efforts.

1.1 Literature Survey

1. Application of Deep Learning in Forest Fire Prediction: A Systematic Re-
view

This paper systematically reviews recent applications of deep learning in forest fire
prediction, analyzing models like CNNs, RNNs, and LSTMs. The authors evaluate
the effectiveness of these models in forecasting fire occurrence, spread, and intensity
based on environmental variables and satellite imagery. The study highlights that deep

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

 AI-Powered Deforestation Detection Using Satellite Images

learning models outperform traditional statistical methods in accuracy and adaptability.

However, challenges remain, such as the scarcity of labeled datasets, varying geographical
conditions, and computational cost. This review contributes by providing a comparative
understanding of the strengths and limitations of existing models, guiding future research
directions in fire risk management.[1]

2. A Learning Strategy for Amazon Deforestation Estimation

This study proposes a novel learning strategy that leverages spatial-temporal satel-
lite image data to estimate deforestation in the Amazon rainforest. The methodology
includes deep learning techniques that integrate remote sensing and environmental vari-
ables. The model aims to improve the timeliness and accuracy of deforestation alerts.
Key findings demonstrate improved performance in detecting early signs of forest degra-
dation. The paper’s contribution lies in its hybrid learning framework, which enhances
scalability and prediction in vast, heterogeneous regions. Limitations include data pre-
processing complexity and model generalization across different terrains. [2]

3. Comparison of Optical and SAR Data for Deforestation Mapping in the

Amazon Rainforest with Fully Convolutional Networks

This research compares the effectiveness of optical imagery and synthetic aperture
radar (SAR) data in deforestation mapping using Fully Convolutional Networks (FCNs).
By training FCNs on both data types, the authors examine their performance under
varying cloud conditions. The study finds that SAR data is more robust in cloudy con-
ditions, while optical data provides better spatial detail in clear skies. The dual-input
model performs best, combining the strengths of both. The paper contributes to improv-
ing year-round monitoring in the Amazon. Limitations include increased computational
requirements and data integration complexity.[3]

4. Deep Learning Tools to Support Deforestation Monitoring

This paper discusses a set of deep learning tools designed for near-real-time defor-
estation monitoring using satellite imagery. It presents a system pipeline that includes
image preprocessing, model inference, and alert generation. Convolutional Neural Net-
works (CNNs) are trained on multispectral satellite data to detect land cover changes.
The study finds that these tools improve detection latency and accuracy, especially
in tropical regions. Contributions include scalable architecture and open-access model
frameworks for environmental agencies. However, challenges such as false positives due
to seasonal vegetation changes remain an area for improvement.[4]

School of Computer Science & Engineering, KLE Technological University, Hubballi - 31. 2
 AI-Powered Deforestation Detection Using Satellite Images

5. Deforestation Detection with Fully Convolutional Networks

This paper explores the application of Fully Convolutional Networks (FCNs) for pixel-
wise classification of satellite images to detect deforestation events. The FCN architec-
ture enables end-to-end segmentation of forest loss areas without requiring hand-crafted
features. The model shows high performance in distinguishing between forest and non-
forest regions over time-series imagery. The study’s strength lies in its automation and
high-resolution capability. Limitations include dependency on well-annotated training
data and potential overfitting in small datasets. It contributes to advancing precision in
forest monitoring systems.[5]

6. Forest Change Detection in Incomplete Satellite Images With Deep Neural

Networks

This research introduces a deep learning approach to handle incomplete or partially

cloudy satellite images for forest change detection. The authors propose a network that
reconstructs missing data using temporal and spatial context and detects changes ac-
cordingly. Results show the model’s robustness against occlusion and data loss. This
innovation supports continuous forest monitoring in areas with frequent cloud cover.
The major contribution is in reducing data gaps using deep image inpainting tech-
niques. Limitations include increased training complexity and sensitivity to temporal
misalignment.[6]

7. Forest Disturbance Detection via Self-Supervised and Transfer Learning With

Sentinel-1 Sentinel-2 Images

This study applies self-supervised and transfer learning methods for forest disturbance
detection using multi-source satellite data (Sentinel-1 SAR and Sentinel-2 optical). The
self-supervised pretraining significantly reduces the need for labeled data while improv-
ing model generalization. The proposed model successfully identifies both abrupt and
gradual disturbances across forested regions. Key findings highlight improved perfor-
mance on small datasets and low-resource regions. The paper’s contribution lies in
its effective integration of modern training paradigms in environmental monitoring. A
noted limitation is the domain gap between pretraining and target data distributions.[7]

8. Deforestation Detection with Fully Convolutional Networks

School of Computer Science & Engineering, KLE Technological University, Hubballi - 31. 3
 AI-Powered Deforestation Detection Using Satellite Images

This paper explores the application of Fully Convolutional Networks (FCNs) for pixel-
wise classification of satellite images to detect deforestation events. The FCN architec-
ture enables end-to-end segmentation of forest loss areas without requiring hand-crafted
features. The model shows high performance in distinguishing between forest and non-
forest regions over time-series imagery. The study’s strength lies in its automation and
high-resolution capability. Limitations include dependency on well-annotated training
data and potential overfitting in small datasets. It contributes to advancing precision in
forest monitoring systems.[8]

9. Deforestation Detection with Fully Convolutional Networks

The paper presents a novel neuroevolutionary approach for classifying deforested areas
in tropical forests. Genetic algorithms are used to evolve neural network architectures
and hyperparameters, adapting to different terrain and data distributions. The results
show better adaptability and competitive accuracy compared to fixed architectures.
Strengths include model flexibility and robustness to noise. The method, however,
requires higher computational resources and extensive tuning. It contributes to the
evolution of automated architecture search in environmental monitoring domains.[9]

10. Online Deforestation Detection

This paper develops an online framework for real-time deforestation detection using
streaming satellite data. The method integrates continual learning with spatiotemporal
filtering to detect abrupt changes in forest cover. It emphasizes minimal delay between
image capture and alert generation. Findings show the system’s effectiveness in timely
interventions. Its contribution lies in bridging the gap between detection and action
in conservation. Limitations include occasional false alarms during natural seasonal
changes and model drift without proper retraining.[10]

1.2 Motivation
The increasing rate of global deforestation poses severe threats to environmental sustain-
ability and climate stability. Traditional monitoring methods rely on manual interpretation
of satellite imagery, which is time-consuming, labor-intensive, and prone to human error.
The delay between deforestation events and their detection often renders conservation efforts
ineffective.
The motivation for this project stems from the urgent need for automated, accurate, and
real-time deforestation monitoring systems that can:

School of Computer Science & Engineering, KLE Technological University, Hubballi - 31. 4
 AI-Powered Deforestation Detection Using Satellite Images

• Provide immediate alerts for significant deforestation events.

• Distinguish between natural forest changes and human-induced deforestation.

• Predict future deforestation trends for proactive conservation planning.

• Support policy-makers with data-driven insights for environmental protection.

• Enable rapid response to illegal logging and environmental threats.

The integration of artificial intelligence with satellite imagery analysis offers unprecedented
opportunities to address these challenges effectively and efficiently.[11]

1.3 Objectives
This report aims to:

1. To automate deforestation detection using satellite images.

2. To develop a web-based application for user interaction and analysis.

3. To classify satellite images into different land cover types (e.g., forested vs. deforested
regions) using AI models.

4. To provide role-based dashboards for stakeholders.

5. To generate detailed analytical reports and alerts.

6. To store and manage uploaded images and prediction results securely using a backend
database (SQLite).

1.4 Problem Defination

The project aim is to develop an AI-powered system that can automatically detect and clas-
sify deforested areas from satellite imagery, enabling faster and more accurate environmental
analysis.

School of Computer Science & Engineering, KLE Technological University, Hubballi - 31. 5
 AI-Powered Deforestation Detection Using Satellite Images

Chapter 2

REQUIREMENT ANALYSIS
This chapter outlines the key requirements necessary for the successful development and de-
ployment of the deforestation detection. It defines the system architecture, essential function-
alities, performance criteria, and database needs to ensure robust and efficient prediction.

2.1 System Model

Figure 2.1: System Architecture

T he image shows the flow from user interaction through a web-based interface, to image pre-
processing, AI/ML classification, and user management, with data exchange between modules
and the deforestation database.

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

 AI-Powered Deforestation Detection Using Satellite Images

The proposed system model illustrates the architecture of an AI/ML-powered deforestation de-
tection framework. This architecture is designed to streamline the process of image-based envi-
ronmental monitoring, particularly focusing on detecting deforestation patterns using machine
learning techniques. The entire architecture is divided into three primary layers—Integration
Layer, Presentation Layer, and Application Layer—each with specific functionalities that work
together seamlessly to achieve accurate and efficient results.

Integration Layer: At the top of the model lies the Integration Layer, which handles the
interaction between the system and the user. This is implemented via a Web-Based User
Interface, which acts as the front end of the application. Users can upload satellite or drone-
captured images through this interface. The web-based GUI is user-friendly, allowing both
technical and non-technical users to engage with the system effectively. It is responsible for
collecting input images, displaying results, managing interactions, and facilitating the user
workflow. The integration layer ensures smooth communication between the user and back-
end processes, sending images to the processing module while also retrieving and displaying
the output, including deforestation analysis and classification results.

Presentation Layer: The Presentation Layer forms the core processing segment of the sys-
tem.
It comprises the following key components:
Image Preprocessing: This module is responsible for refining raw images submitted by the
user. Preprocessing includes noise reduction, image resizing, normalization, and enhancement
techniques to make the images suitable for machine learning analysis. Proper preprocessing
ensures that the images are of a consistent quality and format, which is crucial for accurate
AI/ML classification. This step also helps in reducing computational overhead and improving
the efficiency of the model.
Deforestation Database: The system uses a dedicated database that stores historical image
data, deforestation maps, classification labels, and user activity logs. The database supports
bidirectional data flow—it provides past data for model training and comparison, and it also
stores new analysis results. This database acts as the central repository for system intelligence
and knowledge accumulation, improving over time as more images are processed and classified.

Application Layer: The Application Layer is the foundation of the system and performs
the intelligent decision-making functions.
It consists of the following modules:
AI/ML Classification: This module employs artificial intelligence and machine learning algo-
rithms (e.g., CNNs, Vision Transformers) to classify images into deforested and non-deforested
areas. The model is trained on labeled datasets to recognize visual patterns, terrain types,

School of Computer Science & Engineering, KLE Technological University, Hubballi - 31. 7
 AI-Powered Deforestation Detection Using Satellite Images

vegetation indices, and anomalies. Based on the learned features, it can predict whether a
given image region shows signs of deforestation. The output from this module is then sent
back to the presentation layer for display to the user.
User Management: This module handles user-related operations such as registration, login au-
thentication, access control, and activity tracking. It ensures that only authorized users can
access specific system functionalities and maintains logs for system security and accountabil-
ity. Role-based access control may be implemented to differentiate between administrators,
analysts, and general users.
This modular and layered architecture ensures scalability, maintainability, and efficiency in
real-time deforestation monitoring, making it a suitable choice for both research and deploy-
ment in forest conservation systems.[12]

2.2 Functional Requirements

The system must fulfill the following functional requirements:

1. The system should allow users to register and log in securely.

2. The system should enforce role-based access control.

3. The system should provide functionality to upload satellite images and corresponding
masks.

4. The system should process uploaded images using a trained AI model.

5. The system should generate comprehensive, downloadable reports.

6. The system should present an interactive dashboard with data visualizations.

2.3 Non-Functional Requirements

These are the quality attributes and constraints the system must satisfy:

1. Performance: The system shall return analysis results within a few seconds for moder-
ately sized satellite images to ensure efficient user experience.

2. Scalability: The system shall be designed to support future scalability, allowing it to

handle large volumes of image datasets and integrate with external APIs as needed.

3. Reliability: The system shall operate reliably without crashing or data loss, even under
heavy load during high-volume image processing tasks.

School of Computer Science & Engineering, KLE Technological University, Hubballi - 31. 8
 AI-Powered Deforestation Detection Using Satellite Images

4. Security: The system shall restrict data access based on user roles, enforcing role-based
permissions.

5. Data Integrity: The system shall ensure that all uploaded and processed images are
handled securely without data loss or corruption throughout the workflow.

2.4 Database Requirements

• CustomUser
Fields:
username, email, password – inherited from AbstractUser
role – must be one of: Admin, Researcher, Analyst

• Region
Fields:
name – unique name for the region
region_type – must be one of: Town, Forest Land, Way, Healthy Forest, Homes

• Town
Fields:
name – unique name for each town
region_id – foreign key linking to a Region

• LandType
Fields:
town_id – foreign key linking to a Town
land_type – must be one of: Forest Land, Way, Healthy Forest, Homes

• Result
Fields:
region – name of the region analysed (can be normalized)
prediction – textual description or structured prediction data
image – stored path to the result image

• Image Analysis
Fields:
region – region name (can be normalized)
identifications – text-based/structured tags or AI detections
image – optional uploaded image (nullable)
is_deforested – Boolean flag indicating deforestation presence

School of Computer Science & Engineering, KLE Technological University, Hubballi - 31. 9
 AI-Powered Deforestation Detection Using Satellite Images

Chapter 3

SYSTEM DESIGN
The system design outlines how the deforestation detection system is structured and how
its components interact to achieve the desired functionality. This section details the overall
architecture, data flow, and use cases to provide a comprehensive understanding of the system.

3.1 Architecture Design

Figure 3.1: Architecture Design

This modular architecture ensures scalability, security, and smooth data flow, supporting
the end-to-end functionality from data upload to actionable insight delivery.
The architecture design of the AI-powered deforestation detection system is structured to of-
fer a modular, scalable, and efficient pipeline—from user interaction to data processing and
analysis. This layered and component-based architecture ensures a robust foundation for
processing satellite images, performing deforestation classification, and managing secure user
access, all while maintaining smooth data flow across the system.

User Interaction via Web Browser Interface:

The process initiates with the Web Browser (User Interface), where users interact with the
system. This is the front-facing component that allows users to upload satellite images and
initiate analysis requests. It provides an intuitive interface for both general users and domain

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

 AI-Powered Deforestation Detection Using Satellite Images

experts to input data and view classification results. To maintain system security, user access
is controlled through an Authentication Module. This module validates user credentials and
ensures that only authorized individuals can access the application. It prevents unauthorized
access to sensitive tools and data, maintaining the integrity and confidentiality of the system.

Controller Layer - Django Views:

Once the user is authenticated, requests are managed by Django Views, which function as the
controller layer of the system. Django Views act as the middleware that connects the front-
end interface with the backend modules. It handles the logic for routing user actions—such
as image uploads—to the appropriate components of the system, ensuring a seamless user
experience.
This controller layer also manages session tracking, error handling, and efficient redirection
of input/output across modules, maintaining system responsiveness and stability.

Image Handling and Processing:

When an image is uploaded through the interface, it is first saved temporarily in the Me-
dia Storage module. This module stores raw image files in an organized structure, allowing
the system to access, retrieve, or archive images when needed. Media Storage is crucial for
managing large sets of satellite or drone images, which often require structured indexing and
access control.
Simultaneously, the uploaded image is passed to the Image Processing module, which is the
analytical core of the system. This module performs computational tasks like preprocessing
(e.g., image resizing, normalization, enhancement), and later classification or object detection
using machine learning or deep learning algorithms.
For instance, convolutional neural networks (CNNs) or Vision Transformers might be used
to detect regions affected by deforestation, identify changes in forest cover, or classify land
use based on spectral patterns. This module transforms raw images into meaningful insights,
such as classification maps or deforestation severity levels.

Database and Output Management:

The results of the image processing, along with associated metadata (e.g., timestamps, user
actions, classification outcomes), are stored in the SQLite Database. This structured database
ensures the persistence of both analytical outputs and user-specific information, supporting
future queries, audits, and model training.
SQLite is chosen for its lightweight, file-based architecture, making it an ideal choice for
storing structured data in a compact application. The use of a centralized database enables
efficient data retrieval, report generation, and analysis continuity.

School of Computer Science & Engineering, KLE Technological University, Hubballi - 31. 11
 AI-Powered Deforestation Detection Using Satellite Images

Modular Design and Scalability:

The overall architecture is highly modular, meaning that each component (interface, authen-
tication, processing, storage, and database) functions independently yet integrates smoothly
within the system. This modularity enables scalability—new features or models can be incor-
porated without disrupting the core workflow. [13]

3.2 Data Flow

The data flow diagram outlines the sequential and functional process of detecting deforestation
using satellite or remote sensing data.

Figure 3.2: Data Flow Diagram

Above data flow diagram represents the workflow of an AI-based deforestation detection
system using an external model hosted on Roboflow. The process is broken into several key
stages:
Data Acquisition: This is the starting point of the system where input data is collected. The
data generally consists of satellite imagery, drone footage, or other geo-tagged environmental
images. This step ensures the system receives raw visual data of forest regions to be analyzed.
Preprocessing: Raw images are rarely ready for direct AI processing. In this stage, images
undergo preprocessing steps such as resizing, normalization, noise removal, and sometimes an-
notation. This step ensures uniformity and improves model accuracy during the classification
process.
Model (ic:849ef1 hosted on Roboflow): After preprocessing, the images are fed into a pre-

School of Computer Science & Engineering, KLE Technological University, Hubballi - 31. 12
 AI-Powered Deforestation Detection Using Satellite Images

trained object detection or classification model (referred to as ic:849ef1) hosted on Roboflow.

Roboflow acts as an AI model deployment and inference platform that allows real-time image
analysis using cloud APIs.
AI Classification: This component handles the interpretation of the model’s output. The
AI classifies areas in the image based on the likelihood of deforestation, detecting patterns,
changes in canopy cover, or logging activity. It labels areas as "deforested," "forest," or other
relevant categories.
Impact Assessment: Based on the AI’s classification, this module evaluates the environmental
impact. It calculates metrics such as area affected, rate of deforestation, and severity. This
step transforms raw classification outputs into meaningful environmental insights.
Report Creation: The final stage involves compiling the assessment results into human-
readable reports. These reports can include charts, maps, statistical summaries, and visual
overlays. The goal is to provide researchers, analysts, and decision-makers with actionable
insights for conservation planning or policy enforcement.

3.3 Use Case Diagram

The Use Case Diagram provides a visual representation of the interaction between the system
and its primary user roles: Analyst, Researcher, and Admin. Each of these users plays a
specific and vital role in the end-to-end deforestation detection and analysis process. The
diagram outlines the key functionalities accessed by each user category and showcases how
responsibilities and permissions are distributed across the platform.
1. Analyst Role The Analyst is primarily responsible for the operational and hands-on activ-
ities within the system. Their use cases are focused on direct interaction with the image data
for processing and preliminary analysis.
Their key interactions include:
Upload Image: Analysts can upload satellite images or aerial photographs to the system
for further processing.
Analyze Deforestation: They initiate AI/ML-based analysis on the uploaded images to
detect signs of deforestation or land use changes.
Region Identification: This functionality allows them to mark or identify specific geo-
graphic regions that are affected by deforestation, as determined from the image analysis
results.
This role is critical in providing the initial inputs for the system and in triggering the
primary processing workflows.

School of Computer Science & Engineering, KLE Technological University, Hubballi - 31. 13
 AI-Powered Deforestation Detection Using Satellite Images

Figure 3.3: Use Case Diagram

2. Researcher Role The Researcher role represents users who perform more advanced in-
terpretation and reporting tasks. They depend on the outputs generated by the analyst and
system, and they contribute by analyzing the results for scientific or policy-oriented purposes.
Their core functionalities include:
View Result: Researchers can access and view the analyzed results generated from the
uploaded satellite images. These results typically include classified image data, deforestation
maps, or change detection overlays.
Download Report: This use case enables researchers to download detailed reports, which
may include graphical analysis, statistical summaries, and region-specific observations.
Compare Data: Researchers are also able to compare current results with historical datasets,
allowing for the study of trends, temporal changes, or patterns in deforestation activity.
This role supports deeper insight generation, hypothesis testing, and validation of envi-

School of Computer Science & Engineering, KLE Technological University, Hubballi - 31. 14
 AI-Powered Deforestation Detection Using Satellite Images

ronmental strategies.

3. Admin Role The Admin holds the highest level of system access and control. This role
includes managing backend operations, ensuring data integrity, and overseeing user activities.
The key responsibilities include:
Access Raw Data: Admins can retrieve and review all raw and processed data stored in
the system, including input images and output files.
Compare Data: Like researchers, admins can perform data comparisons, but with broader
access to more datasets and control over which data sources are analyzed.
Download Report: Admins also have permissions to download system-wide reports, either
for auditing, recordkeeping, or institutional review.
The Admin ensures the system runs efficiently and securely, maintaining oversight of both
technical and user-related components.
Collaborative System Roles Together, the Analyst, Researcher, and Admin form a collabora-
tive user ecosystem: Analysts act as data contributors and processors, Researchers generate
insights and conduct evaluations, Admins ensure system integrity and governance.

The use case diagram not only supports technical implementation planning but also aligns with
real-world workflows, making it a practical blueprint for both developers and stakeholders.[14]

School of Computer Science & Engineering, KLE Technological University, Hubballi - 31. 15
 AI-Powered Deforestation Detection Using Satellite Images

Chapter 4

Implementation
This chapter outlines the practical development and integration of the AI-powered defor-
estation detection system. The implementation combines deep learning techniques for image
classification with a user-friendly web interface built using the Django framework. It covers
the tools and technologies employed, the architecture and logic behind the model pipeline, and
how each component—backend, frontend, and controller—was integrated into a cohesive sys-
tem. Emphasis is placed on real-time user interaction, accurate image prediction, and scalable
design using the Model-View-Template (MVT) architecture. The following sections detail the
technological stack, model development process, and Django-based deployment strategy.

4.1 Tools and technologies used

The implementation of the deforestation detection system involves a carefully selected set of
tools and technologies, each serving a critical role across the project pipeline. Python acts as
the core programming language, enabling the integration of both AI models and web frame-
works. For web development, Django is employed, which follows the Model-View-Template
(MVT) architecture and simplifies the creation of a robust and scalable backend. PyTorch, a
popular deep learning library, is used for building and training the machine learning models
responsible for image classification and deforestation detection.
To handle image data, OpenCV and PIL (Pillow) are utilized for tasks like reading, resiz-
ing, and preprocessing satellite imagery. On the frontend, technologies like HTML, CSS, and
Bootstrap are used to design a responsive user interface for uploading images and visualiz-
ing results. For data analysis and manipulation, NumPy and Pandas are integrated, while
Matplotlib supports the visualization of classification outputs and trends. The development
process is supported by tools such as Jupyter Notebook for rapid prototyping and VS Code
for full-stack development. For deployment, SQLite (or Django’s default local database) is
used to store image uploads and prediction results, ensuring lightweight but structured data
management during development or testing phases.[15]

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

 AI-Powered Deforestation Detection Using Satellite Images

Figure 4.1: Comparison of categories with tools and technologies

4.2 Dataset Acquisition and Handling

The dataset is dynamically created through user uploads rather than relying on a predefined
static dataset. Satellite images used for deforestation detection are obtained directly from
users (such as Analysts or Researchers) who upload these images via a web interface.
Steps for Dataset Handling:
1. Image Upload by Users: Users upload satellite images through an upload form on the
frontend. The backend receives the image file through a POST request and stores it securely
on the server.
2. Storage of Uploaded Images: Uploaded images are saved in organized directories within
the server’s media storage, ensuring easy access and management for further processing.
3. Metadata Collection: Along with each image, users submit associated metadata such as
region name and land type classifications. This metadata is saved in the database and linked
to the corresponding image, enabling geographical and contextual analysis.
4. Model Inference on Uploaded Images: The stored images are passed to a pretrained AI
model via an external API (Roboflow). The model analyzes the images and returns classifi-
cation results identifying deforestation and other land types.
5. Saving Results: Prediction outputs along with images and metadata are recorded in the
database, creating a continuously growing dataset that can be used for reporting, visualization,
and further study.

School of Computer Science & Engineering, KLE Technological University, Hubballi - 31. 17
 AI-Powered Deforestation Detection Using Satellite Images

4.3 Model Development and Web Integration

Django Framework Overview:
Django follows the MVT (Model-View-Template) architecture, which cleanly separates the
business logic from presentation and user interaction.

Figure 4.2: Framework of Python Web Development with Django

Below is an explanation of each component shown in the figure 4.2

1. Model (Backend / Database Layer)
Represents the data structure of the application. In this project, it defines the structure for
uploaded images, prediction results, and metadata.
Handles database operations such as storing image paths and prediction outputs. In Project:
A model like
ImageUpload(models.Model) stores uploaded satellite images and model results.
2. View (Controller Logic)
Acts as the intermediary between models and templates.
Processes user requests, applies business logic, and fetches data from models.
Passes data to the templates for rendering. In Project: Views handle file uploads, invoke the
deep learning model for prediction, and return results.
3. Template (Frontend / Presentation Layer)
HTML files that present data to users.
Receive context data from views and render it in a readable, styled format. In Project:
result.html displays prediction results and uploaded images to users.

School of Computer Science & Engineering, KLE Technological University, Hubballi - 31. 18
 AI-Powered Deforestation Detection Using Satellite Images

Chapter 5

Testing
Testing ensures the correctness, reliability, and performance of the deforestation detection
system. Two major testing types were conducted:
Integration Testing – to validate internal component interactions.
Acceptance Testing – to verify that the system meets real-world user expectations.

5.1 Integration Testing

Integration testing is a crucial phase in the software testing life cycle where individual software
modules are combined and tested as a group. The purpose of integration testing is to validate
that the interactions between integrated units or components work as expected. Unlike unit
testing which focuses on testing individual modules, integration testing ensures that the data
exchange and control flow between modules are functioning properly.

In this project, integration testing is conducted to verify the correct communication and
data flow between different components of the system, such as user interfaces, database lay-
ers, application logic, and external APIs. This phase is particularly important for identifying
interface defects, inconsistent data formats, or failure in the interaction between modules that
have been developed and tested separately.

The following key objectives were focused on during integration testing:

Verification of Data Flow: Ensuring that the data passed between modules is accurate and
consistent.
Validation of Functional Interfaces: Confirming that modules interact according to the defined
functional specifications.
Error Handling: Checking how the system behaves when invalid or unexpected data is passed
between modules.
System Behavior Under Real Scenarios: Testing combined modules under real-world usage
scenarios to identify potential integration issues early.

School of Computer Science & Engineering, KLE Technological University, Hubballi - 31 19

 AI-Powered Deforestation Detection Using Satellite Images

Figure 5.1: Test Scenario of Integration Testing

In our system, we tested interactions between the user interface and backend logic, data
processing and storage modules, and third-party services if any. Each integration point was
thoroughly validated through specific test cases which are detailed in the above figure (Figure
5.1).
These test cases ensured that user registration, login authentication, data upload, QR code
generation, and attendance marking were functioning seamlessly in coordination with each
other. The successful execution of these tests indicated that the integrated modules of the
system work in harmony and meet the overall functional requirements.

5.2 Acceptance Testing

Acceptance testing is the final phase of the software testing process, where the complete sys-
tem is tested for acceptability. This type of testing is performed to evaluate the system’s
compliance with the business requirements and to determine whether it is ready for release. It
is typically carried out by end users, clients, or stakeholders to verify that the system behaves
as expected under real-world conditions.
The main goal of acceptance testing is to validate that the developed system fulfills all the
specified requirements and delivers the desired outcomes from the users’ perspective. It en-

School of Computer Science & Engineering, KLE Technological University, Hubballi - 31. 20
 AI-Powered Deforestation Detection Using Satellite Images

Figure 5.2: Test Scenario of Acceptance Testing

sures that all functionalities work correctly in the actual operating environment and that the
user experience aligns with the stakeholders’ expectations.
In this project, acceptance testing focused on real-world scenarios derived from user stories and
requirements. These test cases were designed to simulate how different types of users—such
as analysts, administrators, and collaborators—interact with the system to perform various
tasks.
The acceptance testing process included the following steps:
Requirement Review: Ensuring that each test case directly maps to a functional or non-
functional requirement.
User Role Validation: Testing based on specific user roles to confirm role-based access and
actions.
End-to-End Functionality Checks: Validating complete workflows like registration, image up-
load, prediction generation, report downloads, and email alerts.
Criteria Fulfillment: Checking whether each feature behaves according to the defined accep-
tance criteria.
The outcomes of acceptance testing confirm that the system is stable, user-friendly, and func-
tionally complete. It also signifies that the application is ready to be deployed for actual use
by end users. The above table (Figure 5.2) outlines the key user stories, associated test cases,
and the expected acceptance criteria.

School of Computer Science & Engineering, KLE Technological University, Hubballi - 31. 21
 AI-Powered Deforestation Detection Using Satellite Images

Chapter 6

Results and Discussions

This chapter presents the results obtained from the AI-powered deforestation detection system
using satellite images. The system was evaluated through a series of interactive dashboards
designed for researchers and analysts. The interface includes features for uploading images,
generating reports, forecasting deforestation trends, and analyzing regional data. The func-
tionality and usability of the developed system are demonstrated using screenshots of the
deployed application. These results validate the efficiency of the system in detecting and
analyzing deforestation patterns effectively across different regions.

Figure 6.1: Registration Page

Figure 6.1 shows the registration page, which allows users to create an account. It includes
fields such as Username, Email, Password, and User Type (with a dropdown selection). The
page also includes a prompt for users who already have an account to log in.

School of Computer Science & Engineering, KLE Technological University, Hubballi - 31 22

 AI-Powered Deforestation Detection Using Satellite Images

Figure 6.2: Login Page

Figure 6.2 depicts the login page used by existing users to access the system. It includes
fields for Username and Password, displays messages like successful logout, and provides links
for new registration and password recovery.

Figure 6.3: Research Dashboard - Uploaded Images & Data

Figure 6.3 illustrates the Research Dashboard, where uploaded satellite imagery is displayed
in a card-based layout. Each card shows region-wise deforestation results and offers an option
to download the PDF report for that region.

School of Computer Science & Engineering, KLE Technological University, Hubballi - 31. 23
 AI-Powered Deforestation Detection Using Satellite Images

Figure 6.4: Admin Panel - Dashboard Overview

Figure 6.4 provides a high-level summary for the admin. It shows user counts categorized
by roles, the total number of regions analyzed, and a user management section with options
to edit, delete, or view status.

Figure 6.5: Admin Panel - Researcher Management

Figure 6.5 focuses on the Researcher Management panel. It lists all researchers and allows
the admin to add, activate/deactivate, or delete researcher accounts.

School of Computer Science & Engineering, KLE Technological University, Hubballi - 31. 24
 AI-Powered Deforestation Detection Using Satellite Images

Figure 6.6: Admin Panel - Analyst Management

Figure 6.6 shows the Analyst Management interface. It displays analyst usernames, join
dates, and status. Admins can add or delete analysts, similar to the researcher panel.

Figure 6.7: Analyst Dashboard - Deforestation Trends

Figure 6.7 presents the Analyst Dashboard, used for tracking deforestation patterns. It
contains a table showing region-wise classifications such as “Healthy Forest”, “Houses”, and
“Ways”. The top section summarizes total users, regions, and forests analyzed.

School of Computer Science & Engineering, KLE Technological University, Hubballi - 31. 25
 AI-Powered Deforestation Detection Using Satellite Images

Figure 6.8: Upload and Process Page

Figure 6.8 displays a simple interface that allows users to enter a region name and upload a
satellite image for analysis. This forms the initial step of the deforestation detection workflow.

Figure 6.9: Dashboard with PDF Downloads

Figure 6.9 shows the Research Dashboard in action with multiple classified image cards.
It enables downloading analysis reports in PDF format for regions like Dharwad, Hubli, and
Bijapur.

School of Computer Science & Engineering, KLE Technological University, Hubballi - 31. 26
 AI-Powered Deforestation Detection Using Satellite Images

Figure 6.10: Analyst Dashboard - Deforestation Trends (Tabular View)

Figure 6.10 provides a detailed view of deforestation analysis in a tabular format. Each
row indicates the region and its AI-determined land use category, making trends easier to
analyze.

Figure 6.11: Analyst Dashboard - Forecasting Chart

Figure 6.11 displays a bar chart forecasting deforestation percentages across multiple re-
gions like Munnar, Ooty, and Coorg. The visual layout aids in understanding and interpreting
predicted land use changes.

School of Computer Science & Engineering, KLE Technological University, Hubballi - 31. 27
 AI-Powered Deforestation Detection Using Satellite Images

Figure 6.12: Admin Panel - Report Section

Figure 6.12 shows the report section of the Admin Panel. It presents region-wise satel-
lite images along with AI predictions, such as “Deforestation” or “Healthy Forest”, offering
comprehensive visual insights to the admin.

School of Computer Science & Engineering, KLE Technological University, Hubballi - 31. 28
 AI-Powered Deforestation Detection Using Satellite Images

Chapter 7

Conclusion and Future Scope

7.1 Conclusion
This project demonstrates a practical and efficient system for automated deforestation de-
tection using remote sensing data and machine learning models. The workflow begins with
image acquisition and preprocessing, followed by feature extraction and predictive analysis.
The resulting system can accurately identify deforestation patterns, assisting researchers and
analysts in decision-making and conservation planning. The project’s modular architecture
ensures scalability and adaptability for different forest regions and data sources. The inclusion
of stakeholders like researchers and analysts in the loop further ensures the system’s relevance
and reliability. Ultimately, this project represents a step forward in harnessing AI and data
analytics for environmental protection and sustainable forest management.

7.2 Future Scope

The future scope of this project includes several exciting enhancements. To improve predic-
tion accuracy, the project can incorporate more advanced deep learning architectures, such
as transformers or hybrid CNN models. Integration with GIS data will further strengthen
the analysis by linking deforestation detection with impacts on biodiversity, soil quality, and
hydrology. Additionally, expanding the system to support real-time satellite data feeds can en-
able continuous and automated monitoring of deforestation. A feedback loop for analysts and
researchers can be implemented to refine and enhance the model over time, ensuring it evolves
and adapts to new insights. Cross-regional adaptability is another key area of expansion—by
retraining or fine-tuning models with region-specific data, the system can be made effective
in diverse geographies and forest types. Finally, the project has the potential to influence
policy-making and conservation efforts by collaborating with governmental and environmen-
tal agencies to turn insights from the system into actionable strategies for sustainable forest
management.

School of Computer Science & Engineering, KLE Technological University, Hubballi - 31 29

 AI-Powered Deforestation Detection Using Satellite Images

REFERENCES
[1] D. Lee and Y. Choi. A learning strategy for amazon deforestation estimations using
multi-modal satellite imagery. Remote Sensing, 15(21):5167, 2023.

[2] C. Mambile, S. Kaijage, and J. Leo. Application of deep learning in forest fire prediction:
A systematic review. IEEE Access, 12:190554–190581, 2024.

[3] G. Sartor, M. Salis, S. Pinardi, O. Saracik, and R. Meo. Deep learning tools to support
deforestation monitoring in the ivory coast using sar and optical satellite imagery, 2024.
Preprint.

[4] M. X. Ortega, R. Q. Feitosa, J. D. Bermudez, P. N. Happ, and C. A. De Almeida.

Comparison of optical and sar data for deforestation mapping in the amazon rainforest
with fully convolutional networks. In 2021 IEEE International Geoscience and Remote
Sensing Symposium (IGARSS), pages 3769–3772, Brussels, Belgium, 2021.

[5] D. L. Torres, J. N. Turnes, P. J. Soto Vega, R. Q. Feitosa, D. E. Silva, J. Marcato Junior,

and C. Almeida. Deforestation detection with fully convolutional networks in the amazon
forest from landsat-8 and sentinel-2 images. Remote Sensing, 13:5084, 2021.

[6] S. H. Khan, X. He, F. Porikli, and M. Bennamoun. Forest change detection in incomplete
satellite images with deep neural networks. IEEE Transactions on Geoscience and Remote
Sensing, 55(9):5407–5423, 2017.

[7] P. Helber, B. Bischke, A. Dengel, and D. Borth. Introducing eurosat: A novel dataset and
deep learning benchmark for land use and land cover classification. In IEEE Conference,
2018.

[8] R. S. Kuzu, O. Antropov, C. O. Dumitru, M. Molinier, S. Saha, and X. X. Zhu. Forest

disturbance detection via self-supervised and transfer learning with sentinel-1 & 2 images.
IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing,
17:4751, 2024.

[9] P. B. A. Guilherme, D. B. J. R. Fernanda, Á. Fazenda, and F. A. Faria. Neuroevolution-

based classifiers for deforestation detection in tropical forests. In 2022 35th SIBGRAPI
Conference on Graphics, Patterns and Images (SIBGRAPI), pages 13–18, Natal, Brazil,
2022.

[10] E. Diaz. Online deforestation detection, 2017.

School of Computer Science & Engineering, KLE Technological University, Hubballi - 31 30

 AI-Powered Deforestation Detection Using Satellite Images

[11] G. Fodor and M. V. Conde. Rapid deforestation and burned area detection using deep
multimodal learning on satellite imagery, 2023.

[12] J. Xiang, Z. Zang, X. Tang, M. Zhang, P. Cao, S. Tang, and X. Wang. Rapid forest change
detection using unmanned aerial vehicles and artificial intelligence. Forests, 15:1676, 2024.

[13] M. Ortega, J. Doblas, R. Feitosa, and C. Almeida. Improving deforestation detection

on tropical rainforests using sentinel-1 data and convolutional neural networks. Remote
Sensing, 14:3290, 2022.

[14] U. Ochuba. Treeformers — an exploration of vision transformers for deforestation driver

classification, 2024.

[15] A. Mhatre, N. Mudaliar, M. Narayanan, A. Gurav, A. Nair, and A. Nair. Using deep
learning on satellite images to identify deforestation/afforestation. In Lecture Notes in
Computer Science, 2020.

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