Vivekanand Education Society’s

Institute of Technology
(Affiliated to University of Mumbai, Approved by AICTE & Recognized by Govt. of Maharashtra)

FOREST FIRE DETECTION USING DEEP

LEARNING
Submitted in partial fulfillment of the requirements.
of the degree of

Bachelor of Engineering in Artificial Intelligence and Data Science

by

Ritesh Bhalerao-(5),
Rohit Bhomkar-(8),
Arin Choudhary-(11),
Mukund Tiwari-(62)

Under the guidance of

Supervisor (s):

Guide:- Dr. M. Vijayalakshmi Co-Guide:- Mr. Ajinkya Valanjoo

Department of Artificial Intelligence and Data Science

Vivekanand Education Society’s Institute of Technology
2022-2023

i
 Vivekanand Education Society’s
Institute of Technology
(Affiliated to University of Mumbai, Approved by AICTE & Recognized by Govt. of Maharashtra)

Department of Artificial Intelligence and Data Science

CERTIFICATE

This is to certify that Mr. Ritesh Bhalerao-(5), Mr. Rohit Bhomkar-(8), Mr. Arin
Choudhary-(11), Mr. Mukund Tiwari-(62) of Second Year of Artificial Intelligence
and Data Science studying under the University of Mumbai have satisfactorily
presented the Mini Project entitled FOREST FIRE DETECTION USING DEEP
LEARNING as a part of the MINI-PROJECT for Semester-IV under the guidance of
Guide: Dr. M. Vijayalakshmi and Co-Guide: Mr. Ajinkya Valanjoo in the year
2022-2023.

Date: 25/04/2023

Dr. M. Vijayalakshmi Mr. Ajinkya Valanjoo

Guide Co-Guide

Dr. M. Vijayalakshmi
Head of Department Examiner

ii
 Vivekanand Education Society’s
Institute of Technology
(Affiliated to University of Mumbai, Approved by AICTE & Recognized by Govt. of Maharashtra)

Department of Artificial Intelligence and Data Science

DECLARATION
We, Ritesh Bhalerao, Rohit Bhomkar, Arin Choudhary, Mukund Tiwari from D6AD,
declare that this project represents our ideas in our own words without plagiarism and wherever
others' ideas or words have been included, we have adequately cited and referenced the original
sources.
We also declare that we have adhered to all principles of academic honesty and integrity and
have not misrepresented or fabricated or falsified any idea/data/fact/source in our project work.

We declare that we have maintained a minimum 75% attendance, as per the University of
Mumbai norms.

We understand that any violation of the above will be the cause for disciplinary action by the
Institute.
Yours Faithfully

1. (Ritesh Bhalerao)

2. (Rohit Bhomkar)

3. (Arin Choudhary)

4. (Mukund Tiwari)

iii
 Vivekanand Education Society’s
Institute of Technology
(Affiliated to University of Mumbai, Approved by AICTE & Recognized by Govt. of Maharashtra)

Acknowledgement
On behalf of our group, we express our sincere gratitude and appreciation to our esteemed mentors for
their invaluable guidance, support, and mentorship throughout the duration of this project.

Dr. M. Vijayalakshmi, Ma’am, your expertise and insights have been invaluable in shaping the
direction and scope of this project. Your unwavering support, constructive feedback, and timely
guidance have helped us overcome challenges and navigate through complexities with confidence.

Mr. Ajinkya Valanjoo, Sir, we are deeply grateful for your patience, dedication, and mentorship. Your
extensive knowledge, attention to detail, and commitment to excellence have been instrumental in
enhancing the quality and rigor of our work.

We are also grateful to our college Vivekanand Education Society’s Institute Of Technology, Chembur
for providing us with the opportunity and resources to pursue this project. The support and facilities
extended have been essential in facilitating our research and project work.

Lastly, we extend our heartfelt gratitude to our families for their unwavering support, understanding,
and encouragement. Your belief in us has been a constant source of motivation and inspiration.

iv
 Vivekanand Education Society’s
Institute of Technology
(Affiliated to University of Mumbai, Approved by AICTE & Recognized by Govt. of Maharashtra)

Table of Contents
Abstract
List of Figures

I. Introduction 1
A. Introduction
B. Problem Statement
C. Objectives
D. Scope

II. Literature Survey 5

A. Literature/Techniques studied.
B. Papers/Findings (research papers / white papers / existing systems)

III. Data Set 9

A. Description of data set
B. Data collection methodology

IV. Analysis and Design 12

A. Analysis of the system
B. Proposed Solutions

V. Results and Discussion 15

VI. Conclusion and Future work 18

References 19

v
 Vivekanand Education Society’s
Institute of Technology
(Affiliated to University of Mumbai, Approved by AICTE & Recognized by Govt. of Maharashtra)

ABSTRACT

Forest fires are a growing environmental concern that pose significant threats to ecosystems, wildlife, and
human lives. Detecting forest fires in their preliminary stages is crucial for effective fire management and
mitigation. In recent years, deep learning techniques, such as transfer learning and convolutional neural
networks (CNN), have emerged as promising approaches for accurate and timely forest fire detection.
Transfer learning is a technique that leverages pre-trained neural networks to extract features from images,
which can then be fine-tuned for specific tasks, such as forest fire detection. By utilizing pre-trained
networks, transfer learning allows for the transfer of knowledge learned from one domain to another, even
with limited data. This is particularly useful for forest fire detection, as obtaining large, labeled datasets for
training deep learning models can be challenging.
CNNs, a type of deep learning architecture, are well-suited for image-based tasks, including forest fire
detection. CNNs are designed to automatically learn hierarchical features from images, making them capable
of capturing complex patterns and structures in forest fire images. CNNs consist of multiple layers of
interconnected neurons, including convolutional layers that extract local patterns, pooling layers that reduce
spatial dimensions, and fully connected layers that classify the extracted features.
In recent literature, forest fire detection systems using deep learning techniques have shown promising
results. Transfer learning has been used to leverage pre-trained CNNs, such as VGG, ResNet, and Inception,
for forest fire detection. CNN-based architectures have been proposed, incorporating multiple convolutional
and pooling layers to capture spatial dependencies and improve detection performance.
In conclusion, these techniques enable effective feature extraction and classification from forest fire images,
even with limited data. Further research and development of deep learning-based forest fire detection systems
can contribute to improved fire management and mitigation strategies, helping to protect our valuable forests
and prevent devastating forest fires.

vi
 Vivekanand Education Society’s
Institute of Technology
(Affiliated to University of Mumbai, Approved by AICTE & Recognized by Govt. of Maharashtra)

LIST OF FIGURES

1. Fig 2.1.1 (Convolutional Neural Network)

2. Fig 2.1.2 (Transfer Learning)
3. Fig 2.1.3 (Deep Learning)
4. Fig 5.1 (Detected fire image)
5. Fig 5.2 (Detected no fire image)
6. Fig 5.3 (Detected fire from a video footage)

vii
 Vivekanand Education Society’s
Institute of Technology
(Affiliated to University of Mumbai, Approved by AICTE & Recognized by Govt. of Maharashtra)

INTRODUCTION
1.1 Basic Knowledge

Fire can cause major hazards in this hectic world. All buildings and vehicles used in public transportation
have fire prevention and fire protection systems due to the accelerated number in the fire incidents. Also,
many of the firms conduct a mock fire drill every occurrence of months to protect their employees from the
fire. This would help them to understand what to do or what not to do when a fire situation happens.

Forests are one of the main factors in balancing ecology. It is very harmful when a fire occurs in a forest.
But most of the time, the detection of forest fire happens when it spreads over the aid region. Sometimes, it
could not be possible to stop the fire.

As a result, the damage to the environment is higher than predictable. The emission of large the amount of
carbon dioxide (CO2) from the forest fire damages the environment. As well as it would lead to complete
disappearance of rare species in the world (AL khatib, 2014). Also, it can make an impact on the weather,
and this makes key issues like earthquakes, heavy rains, floods and so on. The forest is a large surface area
filled with trees, lots of dried leaves, woods and so on. These elements encourage the fire when it starts.

The fire can be ignited for many reasons such as elevated temperature in summer seasons, smoking, or some
parties which have fireworks. Once a fire starts, it will remain until it is distinguished completely. The
damage and the cost for distinguishing fire because of forest fire can be reduced when the fire is detected as
early as possible.

1
 Vivekanand Education Society’s
Institute of Technology
(Affiliated to University of Mumbai, Approved by AICTE & Recognized by Govt. of Maharashtra)

1.2. Problem Statement

Fire disasters have posed an ongoing threat to humanity, resulting in widespread destruction and loss of life.
Despite advancements in technology and fire safety measures, accidental fires in residential, commercial,
and industrial buildings continue to occur, leading to severe property damage, economic losses, and loss of
human life.

Additionally, the increasing threat of wildfires, exacerbated by climate change, deforestation, and changing
weather patterns, has further intensified the problem. Wildfires can rapidly spread, destroying vast tracts of
land, displacing wildlife, and posing significant risks to human settlements. Vulnerable populations,
including low-income communities, people with disabilities, and the elderly, are often disproportionately
affected, facing barriers in accessing adequate fire safety measures and challenges in rebuilding and
recovering from fire disasters.

2
 Vivekanand Education Society’s
Institute of Technology
(Affiliated to University of Mumbai, Approved by AICTE & Recognized by Govt. of Maharashtra)

1.3. OBJECTIVES

1. Early Detection: Detect wildfires as early as possible to enable prompt response.

2. Accurate and Reliable Detection: Ensure accurate and reliable identification of wildfires.

3. Wide Area Coverage: Provide comprehensive coverage of forest areas to detect fires in various
locations.

4. Rapid Alert and Notification: Provide timely alerts and notifications to relevant stakeholders
for swift response.

5. Integration with Existing Systems: Integrate with existing fire management and emergency
response systems for coordinated action.

6. Scalability and Adaptability: Design for scalability and adaptability to changing environmental
conditions and technological advancements.

7. Cost-effective and Sustainable: Ensure cost-effectiveness and sustainability in system design

and operation.

3
 Vivekanand Education Society’s
Institute of Technology
(Affiliated to University of Mumbai, Approved by AICTE & Recognized by Govt. of Maharashtra)

1.4. SCOPE

This system can be used by the Government, or any private organization inclined towards environmental
protection and conservation.
The most important objective in fire surveillance is early and reliable detection and localization of the fire.
The data collected through different sensors located at various locations and information will be sent to
Arduino Uno in various places.

In the future, we can develop this model to minimize the energy consumption of all sensors and complete
networks considering the node distribution among clusters using distributed sensing.

4
 Vivekanand Education Society’s
Institute of Technology
(Affiliated to University of Mumbai, Approved by AICTE & Recognized by Govt. of Maharashtra)

LITERATURE SURVEY
2.1 Literature/Techniques studied.

Deep learning, a subset of machine learning, has emerged as a powerful technique for forest fire detection
due to its ability to automatically learn complex patterns from large datasets. Some of the techniques that
have been studied are: -

1. Convolutional Neural Networks (CNNs): CNNs are widely used in image processing tasks, including
forest fire detection. They can automatically learn features from images, such as smoke, flames, or
heat signatures, and classify them as fire or non-fire regions. CNNs can also be used for fire
localization, where they can accurately pinpoint the location of fires within an image or video frame.

Fig 2.1.1(Convolutional Neural Networks)

5
 Vivekanand Education Society’s
Institute of Technology
(Affiliated to University of Mumbai, Approved by AICTE & Recognized by Govt. of Maharashtra)

2. Transfer Learning: Transfer learning is a technique that allows pre-trained deep learning models,
typically trained on large datasets, to be fine-tuned for specific tasks with limited data. Transfer
learning has been employed in forest fire detection systems to leverage pre-trained models, such as
CNNs trained on general image datasets, to achieve good performance even with limited fire-specific
data. This can save training time and resources while improving the accuracy of fire detection
models.

Fig 2.1.2(Transfer Learning)

6
 Vivekanand Education Society’s
Institute of Technology
(Affiliated to University of Mumbai, Approved by AICTE & Recognized by Govt. of Maharashtra)

3. Deep learning: Deep Learning is a subset of machine learning that involves training artificial neural
networks to automatically learn patterns and representations from large amounts of data. The
techniques leverage these deep neural networks to perform tasks such as image recognition, speech
recognition, natural language processing, and more, without explicit programming.

Fig 2.1.3(Deep Learning)

7
 Vivekanand Education Society’s
Institute of Technology
(Affiliated to University of Mumbai, Approved by AICTE & Recognized by Govt. of Maharashtra)

2.2 Papers/Findings

1. "Wildfire Detection and Tracking with Deep Learning" by J. C. S. Jacques Jr. et al. (2019):
This paper proposes a wildfire detection system based on convolutional neural networks
(CNNs) for detecting and tracking fires in real-time using images from visible and infrared
cameras. The authors trained the CNNs on a large dataset of fire images and achieved high
accuracy in fire detection, with low false positive rates.

2. "FireNet: Real-time Wildfire Detection using Convolutional Neural Networks" by M. Uzkent

et al. (2018): This paper introduces FireNet, a real-time wildfire detection system based on
CNNs. The authors trained the CNNs on a large dataset of wildfire images and demonstrated
high accuracy in fire detection, with low false positive rates. The system was also tested on a
field-deployed camera system and showed promising results in real-world conditions.

3. "Forest Fire Detection Using Convolutional Neural Networks" by M. Vakalopoulou et al.

(2018): This paper presents a CNN-based approach for detecting forest fires in high-resolution
satellite images. The authors proposed a novel architecture that combines deep and shallow
CNNs for feature extraction, followed by a fully connected layer for classification. The
proposed model achieved high accuracy in fire detection and outperformed other state-of-the-
art methods.

8
 Vivekanand Education Society’s
Institute of Technology
(Affiliated to University of Mumbai, Approved by AICTE & Recognized by Govt. of Maharashtra)

Data Set
3.1 Description of Dataset

Fire Dataset:
A fire dataset is a collection of data that includes images, videos, or other types of data related to fires,
typically in the context of forest fires or wildfires. It may also include corresponding labels or
annotations indicating the presence or absence of fire. Fire datasets are used for training and evaluating
machine learning or deep learning models for fire detection, prediction, and monitoring tasks. They
provide the necessary data to train models to learn patterns and characteristics of fires from labeled
data, and they are essential for developing accurate and robust fire detection models. Fire datasets can
be collected from various sources and undergo preprocessing, annotation, and augmentation techniques
to enhance their quality and diversity, ensuring that the trained models can effectively detect fires in
different scenarios and conditions.

Non-Fire Dataset:
A non-fire dataset is a collection of data that includes images, videos, or other types of data that do not
contain any instances of fire. It is used as a comparison or contrast dataset during the training and
evaluation of fire detection models. Non-fire datasets are important for training models to distinguish
between fire and non-fire scenes accurately, evaluating model performance, and avoiding false
positives. They typically include images or videos of environments without fire activity and may
undergo preprocessing, annotation, and augmentation techniques to enhance their quality and diversity.

9
 Vivekanand Education Society’s
Institute of Technology
(Affiliated to University of Mumbai, Approved by AICTE & Recognized by Govt. of Maharashtra)

3.2 Data Collection Methodology

1. Data Acquisition: The first step in the data collection methodology is to acquire the relevant data for
training the deep learning model. The images may be in different spectral bands, such as visible,
infrared, thermal, or multispectral, depending on the sensors used in the forest fire detection system.
Meteorological data, such as temperature, humidity, wind speed, and direction, may also be collected
from weather stations or weather models.

2. Data Preprocessing: Once the data is acquired, it needs to be preprocessed to ensure its quality and
suitability for training the model. Data preprocessing may involve various tasks such as image
resizing, normalization, enhancement, and noise reduction. It may also involve filtering out irrelevant
or redundant data and handling missing or corrupted data. Data preprocessing is crucial to ensure
that the data used for training the model is representative, balanced, and of sufficient quality.

3. Data Labeling: Data labeling is the process of annotating the acquired data with relevant labels, such
as fire, smoke, or non-fire, to create a labeled dataset for supervised training of the model. Data
labeling may be done manually by human annotators, or through automated algorithms, depending
on the availability of labeled data and the requirements of the forest fire detection system. Data
labeling is critical for training the model to accurately classify and detect forest fires in the acquired
data.

10
 Vivekanand Education Society’s
Institute of Technology
(Affiliated to University of Mumbai, Approved by AICTE & Recognized by Govt. of Maharashtra)

4. Data Augmentation: Data augmentation is a technique used to artificially increase the diversity and
size of the dataset by generating new samples from the original data. This may involve applying
various image transformations such as rotation, scaling, flipping, or adding noise to the images. Data
augmentation helps in improving the generalization and robustness of the model by exposing it to
different variations of the data.

5. Data Splitting: After data preprocessing and labeling, the dataset is typically split into training,
validation, and testing sets. The training set is used to train the model, the validation set is used to
monitor its performance, and the testing set is used to evaluate the final model's performance. Data
splitting ensures that the model is evaluated on unseen data to assess its generalization and
performance in real-world scenarios.

So, the data collection methodology is a crucial step that directly impacts the accuracy, reliability, and
generalization of the trained model. Careful consideration should be given to the data acquisition,
preprocessing, labeling, augmentation, and splitting steps to ensure that the collected data is representative,
diverse, and of sufficient quality for training a robust and accurate forest fire detection model.

11
 Vivekanand Education Society’s
Institute of Technology
(Affiliated to University of Mumbai, Approved by AICTE & Recognized by Govt. of Maharashtra)

Analysis and Design

4.1 Analysis of the system

Analysis Of Advantages:

1. Accurate and Timely Detection: Deep learning models, when properly trained and optimized,
can achieve high accuracy in detecting forest fires from images. They can capture complex
patterns and representations from large amounts of data, enabling accurate and timely detection
of forest fires in their early stages.

2. Flexibility and Adaptability: Deep learning models can be updated and adapted to changing
environmental conditions, allowing for continuous model improvement and optimization. This
flexibility enables the system to evolve and adapt to new challenges and requirements, making
it adaptable to dynamic forest fire scenarios.

3. Robustness to Variability: Deep learning models, especially with transfer learning, can
generalize well to different environmental conditions, such as varying lighting, weather, and
vegetation, due to their ability to learn hierarchical features from data. This makes the system
robust to variability in forest fire images and enhances its reliability in real-world scenarios.

12
 Vivekanand Education Society’s
Institute of Technology
(Affiliated to University of Mumbai, Approved by AICTE & Recognized by Govt. of Maharashtra)

Analysis Of Challenges:

1. Limited Labeled Data: Deep learning models often require large, labeled datasets for training,
which may be challenging to obtain in the context of forest fires due to the rarity and
unpredictability of fire occurrences. Limited labeled data can potentially affect the model's
accuracy and generalization capability, leading to potential false positives or false negatives in
detection results.

2. Computational Resources: Training and deploying deep learning models can require
significant computational resources, including processing power and memory, which may pose
challenges in resource-constrained environments, such as remote forest areas.

3. Interpretability and Explain ability: Deep learning models are often considered as "black box"
models, lacking interpretability and explain ability. This can be a challenge in understanding
the reasoning behind the model's decisions and gaining stakeholder trust, especially in critical
applications like forest fire detection.

13
 Vivekanand Education Society’s
Institute of Technology
(Affiliated to University of Mumbai, Approved by AICTE & Recognized by Govt. of Maharashtra)

4.2 Proposed Solutions

Some of the proposed solutions for the system could include:

1. Data Fusion: Integrating data from multiple sources, such as satellite imagery, ground-based sensors,
and weather stations, to create a comprehensive and diverse dataset for training the deep learning
model. This would help in capturing various fire-related features, such as smoke, temperature, and
spatial information, to enhance the system's detection capabilities.

2. Edge Computing: Deploying the deep learning model on edge computing devices, such as drones or
IoT devices, to enable real-time inference and reduce dependency on cloud-based resources. Edge
computing can also help in overcoming limitations of unreliable or limited network connectivity in
remote forest areas, making the system more robust and accessible.

3. Visualization and Reporting: Developing a user-friendly visualization and reporting module that
provides real-time visualizations of detected fires on a map interface, along with relevant
information, such as fire location, intensity, and spread direction. This module can also generate
reports and insights on fire patterns, trends, and historical data for better decision-making and
resource allocation.

14
 Vivekanand Education Society’s
Institute of Technology
(Affiliated to University of Mumbai, Approved by AICTE & Recognized by Govt. of Maharashtra)

RESULTS AND DISCUSSION

Fig 5.1 (Detected fire image)

The detected fire image shows that there is a fire present, which appears to be highly intensified as described
in the image analysis. The image also shows surrounding structures that are potentially affected by the fire.

15
 Vivekanand Education Society’s
Institute of Technology
(Affiliated to University of Mumbai, Approved by AICTE & Recognized by Govt. of Maharashtra)

Fig 5.2 (Detected no-fire image)

The detected non-fire image shows that there is no visible evidence of fire present in the image. The image
depicts the structure of a building that was captured. The absence of fire in the image may be relevant in
certain contexts.

16
 Vivekanand Education Society’s
Institute of Technology
(Affiliated to University of Mumbai, Approved by AICTE & Recognized by Govt. of Maharashtra)

Fig 5.3 (Detected fire from a video)

The video footage of the fire shows that there is an ongoing fire, which appears to be highly intensified. The
video can provide valuable information for emergency responders and can help with the management of the
fire.

17
 Vivekanand Education Society’s
Institute of Technology
(Affiliated to University of Mumbai, Approved by AICTE & Recognized by Govt. of Maharashtra)

Conclusion & Future Work

It is easier to suppress a fire in its starting stage than in the later stages. Hence, the important goal is quick,
reliable detection and localization of the fire. The proposed fire alert system overcomes the need of human
intervention to continuously monitor the forest area.

Monitoring and detecting are done by the sensors installed and message alerts are used to alert the required
authorities. It gives instantaneous information on the location of the fire to the authorities and helps block
the fire before it reaches other sites.

This fire alert system is power efficient, low cost and low maintenance, and the equipment is durable and
reliable.

In the future, we can install a wind sensor to the system which helps to determine the direction of the fire
and the rate at which it will spread.

18
 Vivekanand Education Society’s
Institute of Technology
(Affiliated to University of Mumbai, Approved by AICTE & Recognized by Govt. of Maharashtra)

REFERENCES

1. https://pyimagesearch.com/2019/11/18/fire-and-smoke-detection-with- koras-and-
deep-learning/

2. https://towardsdatascience.com/early-fire-detection-system-using-deep- learning-and-
opencv-6cb60260d54a

3. https://www.researchgate.net/publication/357811123_FOREST_FIRE_DETECTION
_USING_MACHINE_LEARNING

4. Saponara, S., Elhanashi, A. & Gagliardi, A. Real-time video fire/smoke detection based
on CNN in antifire surveillance systems. J Real-Time ImageProc 18, 889–900 (2021).

5. A Arul, R. S. (2021, May). Fire Detection System Using Machine Learning.

RetrievedfromresearchGate:https://www.researchgate.net/publication/351926970_Fir
e_Detection_System _Using_Mac hine_Learning.

19