INTRODUCTION

1.1 PROJECT OVERVIEW

Our project is a system based on upon detection of forest fires or wildfires using state of the art
machine learning techniques. The probability of initiation of a forest fire can be viewed by the
authorities on a mobile application, so that the area can be checked for fire by deploying drones.

The drones will be equipped with sensors, Raspberry pi 3 , neural stick, APM 2.5,GPS,Wifi .
The APM will help to drone for flight control. The neural stick will be used for real time image
processing using our state of the art deep learning model. And as soon as forest fire is detected
the UAV will send an alert message to the concerned authority on the mobile App along with
location coordinates of the fire, image and the amount of area in which forest is spread using a
mesh messaging.

We developed a platform that is completely automated since both drones have on-board
computers and processing capabilities. They can detect the fires based on the image capture
without the need for centralized computing engine.

1.1.1 Technical Terminology

1. Use of a small-scale of convolution neural networks is implemented with the help of YOLOv3.
2. Cross Platform Mobile application would send an alert notification to the concerned
authorities.
3. Drone will be occupied with onboard computer(neural stick) that will provide real time image
processing for forest fire detection.

1.1.2 Problem Statement

Forest fire is a global environmental problem causing extensive damage every year. According to
the International Union for Conservation of Nature (IUCN) Report “Global Review of Forest
Fire 2000,” wildfire is a natural phenomenon. However, over 90% of all wild land fires are due
to human action causing significant forest loss, that is 6–14 million hectares of forest per annum,

1
and 30% of the CO2 in the atmosphere produced from forest fires. This leads to enormous
economic losses, damage to environmental, recreational and amenity assets, global warming and
loss of life. There is a strong recognition that action is needed to catalyze a strategic international
response to forest fires.

Recently, a record-breaking fire was seen ripping through the Amazon and Australia, which
caused great damage to the ecosystem and contributed to the increase in air pollution and cause a
death of almost 480 million animals. Even though it could have been easily prevented by the
authorities, due to lack of proper surveillance and human negligence it worsened.

1.1.3 Goal

The most important factors in the fight against forest fires include the earliest possible detection
of the fire event, the proper categorization of the fire and fast response from the fire fighting
departments.

We aim to develop an automatic early warning system integrating multiple sensors and state of
the art machine learning algorithm which has a minimum number of false positives and give a
good accuracy in real time data and in the lowest cost possible to our drone to monitor Forest fire
as early as possible and report it to the concerned authority.

So that immediate action will be taken to stop it from spreading and causing loss of millions of
lives and money.

We will take advantage of recent advances in multi-sensor surveillance technologies by

employing both machine learning and infrared cameras to monitor the site and the surrounding
area.

1.1.4 Solution

The work was started by discussing the techniques to detect the presence of fire for our
application and finally the use of a small-scale of convolution neural networks is implemented
with the help of YOLOv3 was decided upon by seeing the computational speed i.e no. of frames
process each second and accuracy.

2
Secondly auto-piloted drone is implemented using APM 2.5.ArduPilot is an open source,
unmanned vehicle Autopilot Software Suite,[1] capable of controlling autonomous.

Thirdly, our cross platform native mobile application was chosen to be built by using flutter.

Fourth, for communication between drone and concerned authority we choose a network of
LAN.

1.2 NEED ANALYSIS

Wildfires were one of the main causes of archeological and cultural heritage sites being
destroyed, especially in the Mediterranean region. Such sites, treasured and maintained for long
periods of time, are usually surrounded by ancient and precious vegetation or are located near
forest areas. The increase in seasonal temperatures has resulted in a spike in the number of self-
ignited fires in forest areas that are catastrophic when fanned by winds and fuelled by dry
vegetation. Moreover, arson events have been reported on several occasions, while unintended
fires are commonly caused by human carelessness and lightning strikes. Extreme weather
conditions such as storms or floods have also posed great risks for these sites.Significant damage
has been caused by similar fires in archeological areas and treasures throughout the
Mediterranean, especially during the summer months. In addition to taking precautionary
measures to avoid forest fires, the only way to avoid human casualties and damage to the
environment and cultural heritage has been early warning and immediate response to a fire break
out. Fast and accurate fire detection and localization has become the most important objective in
fire monitoring, because it is much easier to contain a fire when the location of the ignition point
is established and the fire is at an early stage. An automatic fire detection system that relies on
multi-sensor networks should be able to provide early fire warning as well as collect location and
fire spread information to facilitate effective fire management. Based on this information,
firefighters can be directed on the objective of containing the fire before it reaches cultural
heritage sites and quickly eliminate it by using the fire appropriate equipment and vehicles.
Because of the lack of an integrated approach, most commercial wildfire surveillance systems do
not realize the full potential of available technologies. Most systems use visible range cameras
mounted on watch towers to monitor large forest areas. Some systems use infrared cameras,
which are usually much more expensive than regular pan-tilt-zoom cameras, and may be limited

3
in their range. Few systems use wireless sensor networks that can provide real-time feedback for
fire detection and evolution. All of the above approaches have their own limitations and
advantages. This project now aims to develop an automatic early warning system that integrates
multiple sensors to remotely monitor archaeological and cultural areas of interest for the risk of
fire and extreme weather.

1.3 RESEARCH GAP

The major gap is the major difficulty in wildfire image classification is the lack of unified
identification marks, the fire features of color, shape, texture (smoke, flame, or both) and
background can vary significantly from one scene to another. To overcome this problem we
come up with deep learning model that removes our this difficulty.

The main issue of this UAV monitoring system is that the whole system is operated by observers
in surveillance center. Limited by the wireless transmission distance or the blind zone of
GPRS/GSM network, real-time video transmission from the UAV can hardly be guaranteed. The
surveyors often wait for the return of the UAV and examine its stored video. It is time
consuming, labor intensive, and inefficient. Thus, an autonomous fire localization and
identification system for this UAV is right now the top demand. Early reported works on fire
detection are generally based on sensor techniques. For example, heat and smoke detectors are
mostly employed. These methods work properly in indoor environments. However, when applied
to large outdoor areas, environmental factors can significantly impact their performance. Later,
sensor network-based methods are reported effective in prediction rather than detection of
wildfire, this is mainly because these methods depend on restricted data of relative temperature
rise or wind speed in calculating the probability and intensity of fire. However, due to the limited
measuring distance of each sensor, fine grained wildfire mapping area demands very dense
deployment of sensors. It is difficult to implement in practice. Advanced sensor technology has
been applied in fire detection: infrared sensors are used to capture the thermal radiation of fire,
light detection and ranging system (LIDAR) is employed to detect smoke by examining the
backscattered laser light. Yet, these optical systems are also found sensitive to the varying
atmospheric and environmental conditions: clouds, light reflections, and dust particles may lead
to false alarms.

4
Satellite imagery is a common method for wildfire detection , but the long scan period and low
flexibility make it difficult for early fire detection. Infrared (IR) thermo graphic cameras are used
to generate thermal image of an area. It can obtain reliable heat distribution data for fire
detection. Yet most aerial IR thermo graphic imaging systems work on the wave band from 0.75
to 100 µm. They detect much less environmental information on this band. This information can
also be very important, especially when flammable and combustible materials are presented.

1.4 PROBLEM DEFINITION AND SCOPE

Wildfire is a natural disaster, causing irreparable damage to the local ecosystem. Sudden and
uncontrollable wildfires can be a real threat to residents’ lives. Statistics from the National
Interagency Fire Center (NIFC) in the USA show that the burned area doubled from 1990 to
2015 in the USA. Recent wildfires in Amazon Rainforest and in Australia (reported by CNN)
have already resulted in more than 40 deaths and 50 missing and more than 480 million animals
are dead . More than 200,000 local residents have been evacuated under emergency. The
wildfires occur 220,000 times per year globally, the annual burned area is over 6 million
hectares. Accurate and early detection of wildfire is therefore of great importance.

The scope of the project is to make to develop an automatic early warning system integrating
multiple sensors and state of the art machine learning algorithm which has a minimum number
of false positives and give a good accuracy in real time data and in the lowest cost possible to our
drone to monitor Forest fire as early as possible and report it to the concerned authority.

1.5 ASSUMPTIONS AND CONSTRAINTS

1) Since this project is based on real time image analysis so it will require internet in forest for
data transmission to alert the concerned authorities about forest fire.

2) At the time of forest fire, wind speed should be normal so that drone could fly normally and
data transmission will be smooth.

3) Camera fitted to drone should be of high resolution.

5
4) Since this project is based on an app, so the concerned authority should have mobile of
required specification.

5) The dedicated path of drone should be clear there should be no obstacle.

1.6 APPROVED OBJECTIVE

Following objectives have been identified:

1) Create a machine/deep learning algorithm for detection for wildfire in real time.

2) Create a drone with a camera and raspberry pie mounted and an on-board computer(neural
stick) on it for real time video analysis

3) Finding the geo location of the fire.

4) Create a cross platform native mobile application for the authorities to get notification.

5) Creating a network in forest to send notification from drone to concerned authorities.

6) Find the nearest fire fighting station from the detected region of forest fire.

1.7 METHODOLOGY USED

The objectives of the project listed above will be achieved in small steps as per the schedule and
plan.
● We use a drone equipped with a high resolution camera, a mini processor and ArduPilot
Mega for the flight controller
● There is a barometer, Global Positioning Sensor (GPS), an inertial measurement unit and
compass. The GPS and compass are used as a navigation system. The barometer

6
 measures the air pressure that is used as the reference to maintain the height of the UAV.
The IMU consists of accelerometer and gyroscope sensors that are used to estimate the
vehicle’s position.
● The data from both the high resolution camera and GPS are processed by the Raspberry
Pi 3, which serves as a mini processor. The Raspberry Pi 3 sends data to the web server to
be accessible online and real-time app.
From the previous available data available we will predict the region and time, when will forest
fire will happen.
For auto-piloted drone we decided to choose APM 2.5. We can manually select the region where
we want our drone to move continuously in a dedicated region.
We embed our yolov3 model in neural stick, As soon as image is captured it is processed and
check for forest fire. And if detected more than a confidence level it will send notification.
The results of the data processing are sent to the concerned authority using raspberry pi3, the
data transmission used the Transmission Control Protocol system.

1.8 PROJECT OUTCOME AND DELIVERABLE

A neural stick raspberry pie and a camera module are to be purchased. Our end user is:

1. Authorities

For authorities with their smartphone with our application being installed before-hand,
they can get to no know about the wildfires.

1.9 NOVELTY OF WORK

The novelty of our solution lies in the fact that our system is working on the very simple fact,

Getting notification and locating the point of forest fire is easy using the app, As the drone will

Send the Geo-location of the fire to the app. And Our model is every efficient from the

7
Traditional style of detecting forest fire like using watch towers as it required many humans
power and time also.

8
 REQUIREMENT ANALYSIS

2.1 LITERATURE SURVEY

2.1.1 Theory Associated With Problem Area

Early detection of fire and smoke is vital to encourage action in order to avoid large-scale
damage. Many traditional methods of fire and smoke detection, which are supposed to detect the
presence of smoke or flames, are using sensor-based instruments. A main disadvantage is
identification of fires and flames only in the area of construction sites. Forests are an important
component in preserving nature's balance. The forests cover about one-third of the surface of the
Earth and are home to two-thirds of the earth's species. Forests are the Earth's source of
biodiversity. Unfortunately, forest fires have burned millions of hectares of land every year, and
hundreds of millions of dollars are spent on extinguishing such fires. There are a number of
solutions to address this issue that have been developed. Early fire detection systems in the forest
are one of the ways used to prevent forest fires. Massive efforts were made to track, locate, and
rapidly extinguish the fire before it becomes too high. Traditional fire control and detection
methods use people to track forests unwaveringly. This method, however, requires a high cost
and is not safe for people. Recently, remote sensing systems have become one of the most
efficient methods for forest monitoring. The advancement of electronics, computer science, and
digital camera technology has allowed the monitoring and detection of forest fire computer-
based remote sensing systems. For forest fire tracking and detection, remote sensing solutions
can be divided into three groups: land-based systems, manned air-based systems, and satellite-
based systems. Nevertheless, these systems contain a number of technological issues and
procedures. Manned aircraft are usually large and costly. Furthermore, the safety of the pilot is
also a part to consider. Unmanned aerial vehicles are becoming an increasingly attractive and
realistic option with computer-based remote sensing systems. UAVs are also relatively cheaper
to continuously monitor and detect forest fires in addition to being faster and more mobile.
UAVs can also be combined with remote sensing techniques to add value to
existingsolution.Therefore, UAVs can operate in dangerous environments that can not be reache
d safely by humans. These are the reasons why UAVs have become one of the solutions that dra
w the attention of the world to combat forest fires.

9
2.1.2 Existing Systems and Solutions

Satellite-Based Systems

Earth-orbiting satellites and even air-floating instruments were used for forest fire observation
and detection. Satellite images collected by two main satellites launched for forest fire detection
purposes, the advanced high-resolution radiometer 2, and the spectroradiometer of moderate
resolution. Sadly, these satellites can provide pictures of the Earth's regions every two days,
which is a long time for fire scanning; weather conditions can affect the quality of satellite
images as well.

Optical Sensor and Digital Camera

There are actually two different types of sensor networks for fire detection, camera tracking
and wireless sensor network. The advent of sensors, digital camera, image processing, and
industrial computers led to the development of an electronic, automatic early detection and
forest fire warning system.

Different types of detection sensors can be used in terrestrial systems :

 Video camera, sensitive to visible smoke spectrum recognizable during the day and fire
identifiable at night
 .infrared thermal imaging cameras focused on fire heat flow detection
 IR spectrometers to recognize smoke spectrum characteristics
 light detection and range systems — measuring laser rays emitted from smoke particles.

Wireless Sensor Networks

The line of sight and the early stage of the fire process problem could be solved with the
second type of sensors. A new technology called wireless sensor network is nowadays
receiving more attention and has started to be applied in forest fire detection. The wireless
nodes integrate on the same printed circuit board, the sensors, the data processing, and the
wireless transceiver and they all consume power from the same source batteries. Unlike cell

10
 phones, WSN do not have the capability of periodic recharging. The sensors are devices
capable of sensing their environment and computing data. The sensors sense temperature,
pressure and humidness, moreover as chemical parameters like carbon monoxide
gas, dioxide, and gas. The sensors operate during a self-healing and self-organising wireless
networking setting. One sort of wireless technology is ZigBee that may be a new
industrial commonplace supported IEEE 802.15.4. This technology emphasises
low value battery powered application and little star panels and is suited
to low information rates and little vary communications. Wireless device networks have
seen fast developments during a sizable amount of applications.

2.1.3 Research Findings for Existing Literature

Table 1: List of various Research Papers

Tools /
Paper Title Findings
Technology

DNN’s are difficult to train so they provide empirical

evidence showing that residual networks are easier to
Deep Residual Learning for
Yolo optimize and gain accuracy.
Image Recognition

Here they use neural networks to identify key seven human

emotions: anger, disgust, fear, happiness, sadness, surprise
Recognizing Facial Expression
R-CNN and neutrality.
using deep learning

It represents a residual learning framework to ease the

training of networks that are substantially deeper than those
Deep Residual Learning for Resnet, VGG,
Image Recognition Google Net used previously.

11
2.1.4 The Problem That Has Been Identified

In order to find a fireplace, the traditional GPS fails to grant the correct results .GPS has been
used and been remarkably nice in terms of locating the
thing, however it's significantly didn't offer correct results for the
locating fire. Thus there's vast want for Associate in Nursing correct fire detection
system that overcomes the barriers moon-faced by the GPS.

2.1.5 Survey of Tools and Technologies Used

1) NASA’s Fire Information for Resource Management System

(FIRMS) distributes Near Real-Time (NRT) active fire data within 3 hours of satellite

overpass from NASA’s Moderate Resolution Imaging Spectroradiometer (MODIS)

and NASA’s Visible Infrared Imaging Radiometer Suite (VIIRS).

2) Another system which is popular round the world uses motion and color clue features
combined with Edge blurring and flicker features, to detect fire and flame

3) A chromatically-based decision rule and diffusion dynamic decision

Rule is used to draw conclusions about whether each pixel does or does not represent

Smoke.

4) Computer vision based measurement method which uses augmented reality to measure the
dimensions of the flame.

5) el at Zhang proposed two joined deep CNNs to detect fire in forest images. In Whole image is
tested by a global image classifier; then if a Fire is detected in any region, another classifier is
used to identify the exact location of the fire in that image.

6) A deep CNN model for smoke detection inspired by AlexNet.

12
Conclusion: variety of smoke and fireplace image classification approaches are
proposed to beat this problem; but, most of them trust either rule-based
methods or on handcrafted options. a completely unique deep convolutional neural network
algorithm to realize high-accuracy fireplace and smoke image detection is projected by United
States.

Instead of mistreatment ancient corrected linear units or tangent functions, adaptive piecewise
linear units within the hidden layers of the network mistreatment CNN, is used .This method is
capable of classifying each smoke and fireplace pictures at a similar time, and
many benefits are offered and higher performance is exhibited than alternative existing visual
recognition CNN models for the popularity of fireplace and smoke in pictures.

2.2 STANDARDS

 IEEE standard 830

IEEE standard 830 governs a set of rules for building the software requirement.Specifications
and also offers a demo through to making a good SRS.Since we are building the SRS document,
hence we are incorporating this standard.

 Industry 4.0
It governs the current trend of automation and information exchange.

 ISO/IEC 9241-11
Part 11 of this standard deals with the extent to which a product can be used by specified users to
achieve specific goals with satisfaction in a specified context of use. This standard projected a
framework that describes the usability elements and therefore the relationship between them.
In this standard, the usability is taken into account in terms of user performance and satisfaction.

2.3.1 Introduction
2.3.1.1 Purpose

13
 To develop a system that will facilitate in detecting forest or bush fire at a budding
stage in order to prevent mass mislaying of green resources.
2.3.1.2 Intended Audience and Reading Suggestions
For the implementation of this project, Fire disaster authorities and forest departments
will be our target audience for all kinds of implementation and testing purposes. As a
reading suggestion, we suggest the reader’s to go through the various techniques
which have been proposed to detect this problem at an early stage.
2.3.1.3 Project Scope
The project scope is as follows :-

 Aim to provide an automated forest fire detection system.

 To simplify the process of locating a forest or a bush fire at a germinating
stage in order to prevent mass destruction of flora and fauna.
 To offer an ease of access to a fire department in locating forest fire if any.
 To mitigate the response time of concerned authorities in case of any fire
incident.
2.3.2 Overall Description
2.3.2.1 Product Perspective
One of the main causes of destruction of archaeological and cultural heritage sites,
especially in the Mediterranean region, is wildfires. These sites, treasured and tended
for long periods of time, are usually surrounded by old and valuable vegetation or
situated close to forest regions. The increase in seasonal temperatures has caused an
explosion in the number of self-ignited fires in forested areas, which fanned by winds
and fuelled by dry vegetation become disastrous. Extreme weather conditions such as
storms or floods also pose greats risk for these sites. Beyond taking precautionary
measures to avoid forest fires, early warning and immediate response to a fire break
out is the only way to avoid human losses and environmental and cultural heritage
damage. Although several technologies based on different sensors have been
proposed for wildfire surveillance, the majority of existing fire detection systems does
not realize the full potential of state-of-the-art technologies due to the lack of an
integrated approach. In this project, an automatic early warning system integrating

14
 multiple sensors to remotely monitor areas of archaeological and cultural interest for
the risk of fire and extreme weather conditions was developed. The system integrates
various sensors including optical cameras, infrared cameras at different wavebands,
passive infrared (PIR) sensors, a wireless sensor network of temperature and humidity
sensors and local weather stations on the deployment sites. The signals and
measurements collected from these sensors are transmitted to the control centre,
which employs intelligent computer vision and deep learning to combine sensor
information and detect the presence of fire or smoke. The control centre is capable of
generating automatic warning signals for smoke/flame detection and abrupt
temperature rise.
2.3.2.2 Product Features
1 Continous monitoring of areas which are highly vulnerable to forest or bush fire.
2 Saves Human labour,effort and intelligence from doing repeated and endless task.
3 Eliminates Human errors and negligence.
4 Monitoring areas where human survival is next to impossible.
5 Detection survivors, if any in an avalanche or a air plane crash with minor changes in the
code.
6 Innovative system which can decrease the response time of variuos forest fire
departments.

2.3.3 External Interface Requirements

2.3.3.1 User Interfaces
GUI along with meaningful framesin association with google maps which depict the
exact location of the region where fire has detected by the unmanned aerial vehicle.

2.3.3.2 Hardware Interfaces

• Drone kit APM 2.5
• Raspberry Pi 3
• Neural stick
• Camera

15
 • Infrared sensor
• Node MCU
• Smoke sensor
• IMST IC 880A
2.3.3.3 Software Interfaces
Front End: Flutter app, HTML and Js for web page.

Back End: PHP

2.3.4 Other Non-functional Requirements

Non-functional requirements are set of requirements with specific criteria to judge the
systems operation. They cover ease of use to the client, security, support availability,
operational speed, and implementation considerations.
2.3.4.1 Performance Requirements
1. The speed of locating the desired entity should be fast.
2. The accuracy of the result should be good.
3. App size should be small.
2.3.4.2 Safety Requirements
1. App should be compatible on various operating systems.
2. The received signal strength values should be reliable.
3. The app should be compatible with the various android versions.
2.3.4.3 Security Requirements
1. The system should be very secure as it contains information about areas which are
vulnerable to forest fires.

The server data should be secured as it contains the info

2.4 COST ANALYSIS

2. rmation about the geography of various forest spreads.

16
 3. For a small organization like in this case our college, we would store data on a
local server instead of a cloud to prevent it from external attacks.

Table 2 : Cost Analysis Detail

Component Price
Raspberry-Pi 4,525

Drone Kit 8,299

Node MCU 269

Camera Module 1,500
Humidity Sensor(BME280) 800
Smoke Sensor 1,500
Neural Stick 8,500
SIM800A Module 825
Total(Estimated) 26,218

2.5 RISK ANALYSIS

Though our project is taken care of every causality but certainly there can be many flaws
and further risks involved if not checked every part of our project's functioning. Some of the
major flaws can be :-
1. System can alarm the fire authorities because it may not be able to distinguish between a
forest fire and and bon-fire used by travelers for their survival.
2. Unavailability of good wifi signal strength in the area of interest.
3. Unmanned aerial vehicles may get attacked by any wild animal.
4. Monitoring drone can get stuck in dense bushes.

17
5. Database collection is a vast process and their might be some errors in data collection
which may lead to error.

18
 METHODOLOGY ADOPTED

3.1 INVESTIGATIVE TECHNIQUES

The very first investigation we conducted was to identify the technologies required according to
the Need Analysis. First of all the decision taken was on the mobile client, we thought of using
Flutter for making the mobile client, since Flutter is a cross platform application development
frame work. For the backend, we figured out to use Firebase as our data structure framework.
Rest programming of backend is done on Flask.

Table 3 : Investigative Techniques

Investigative Projects Investigative Techniques

S. No.
Techniques Description

We have to predict most difficult

task ie.(Forest fire detection
1. Descriptive
techniques). So we read some
research papers on it.

We applied different machine

2. Comparative learning models to predict the best
output.

Best Model was CNN, Hyper

3. Experimental
parameter tuning

19
3.2 PROPOSED SOLUTION

The proposed flame identification methodology as follows-

• Burning substances principally identified as hopeful flame areas utilizing a histogram-based

division strategy to expel the non-fire foundation.

• The old-style optical stream strategy is then connected to identify moving articles for disposing
of stationary non-fire objects in the candidate flame areas.

• Next, the movement vectors determined by the optical stream are additionally examined to
decrease false alarm rates brought about by hot moving items.

• After that, the converged applicant fire locations of visual and IR pictures are sectioned; a
definitive flame affirmation depends on these pixels in the convergence areas.

• Finally, when the flame areas affirmed, fire zones are followed by mass counter-strategy.

Figure 1 : Proposed Solution block Diagram

20
3.3 WORK BREAKDOWN STRUCTURE

Figure 2 : Work Breakdown Structure

21
3.4 TOOLS AND TECHNOLOGIES USED

Deep learning and Computer Vision:

 Python(3.x)
 Tensorflow(version2)
 OpenCV
 Keras
 Numpy
 Mathplotlib
 Scikitlearn
 Matlab
 Hardware Implementation:
 Raspberry Pi4
 Hardware interfacing

1. Front End

Cross platform Mobile Application - Flutter

2. Back End

Database - RoboMongo for MongoDB (Free of Cost)

Rest API - Atom for Node.js (Free of Cost)
 Database - Firebase
 Rest API - Flask

22
 DESIGN SPECIFICATIONS

4.1 SYSTEM ARCHITECTURE

Figure 3 : System Architecture

23
4.2 DESIGN LEVEL DIAGRAMS

Figure 4 : Design Level Diagram

24
4.3 USER INTERFACE DIAGRAMS

Figure 5:Snapshot of the Mobile App

25
4.4 SNAPSHOTS OF WORKING PROTOTYPE MODEL

26
 CONCLUSIONS AND FUTURE DIRECTIONS

After performing a detailed experiment, we have arrived at:

5.1 WORK ACCOMPLISHED

With this project the objective which we have accomplished are as follows:-

● So far we have trained our model on more than 5000+ images to predict fire. Secondly
our YOLO based forest fire detection system is working with the accuracy of more than 90 % .
● A basic structure of the mobile app and web interface is ready.
● We have researched and deeply analyzed more than 20+ research papers to find the faults
in the prediction method
● We can update the authorities via sms by using our current system .
● We tested our model on my videos from social media .
● We have created a web page which uses NASA satellite data for prediction the vulnerable
states in India towards forest fire .

● Continuous monitoring of areas which are highly vulnerable to forest or bush fire.
● Saved Human labour effort and intelligence from doing repeated and endless task.
● Eliminated Human errors.
● Developed a system which can monitor areas where human survival is next to impossible.
● Developed a system which also be used to detect survivors, if any in an avalanche or a air
plane crash with minor changes in the code.
● Innovated a system which can decrease the response time of variuos forest fire
departments.
● Early detection of bush fire to prevent any large scale destruction.
● Lastly, used technology as a whistleblower to alarm concerned authorities and prevent
massive environmental mislaying.

27
5.2 CONCLUSIONS

The technology used by us to locate a forest or a bush fire is solely based on the concept of
Image division using the technique of Thresholding using an unmanned aerial vehicle. We
have been able to test the code in our nearby surroundings and have successfully inferred the
results which are quite accurate. Considering this first step done, we now aim to make an
android app where we would deploy the technology used above in the forest fire detection
system. All in all, this app will help the forest departments of various vastly spread forest in
early detection of forest fire and will ease the work of those departments which are endlessly
monitoring the fire prone forest round the globe.

5.3 ENVIRONMENTAL (/ ECONOMIC/ SOCIAL) BENEFITS

Fires threaten forests which results in enormous material and environmental damage. Protection
of forests against fire is based on a variety of preventive measures and measures for fighting
against forest fires, in order to minimize the total damage. In addition to other preventive
measures, early detection and fire extinguishing in the initial stage are important in the protection
of forests against biotic and abiotic factors. The existing surveillance of forest areas is unreliable
and inefficient; therefore, forest fires are a serious threat to the development of forestry.
Forest fire detection can be very useful in many areas to ease the tedious task of monitoring
vastly spread forests and also this project can be helpful for environmental benefits. One can
deploy this system for different purpose like monitoring,surveillance and guard those areas of
forest stretch where human reach is next to negligible.

28
5.4 FUTURE WORK PLAN

We have laid down some important points which define the scope of our project and we wish to
follow these points in order to build a system which will truly cater to the needs of forest and
environmental authorities with no errors.
The points are:
1. We need to apply a minimalistic approach in redefining the UI of the mobile application used
by authorities to create a low latency prediction process, thereby effectively decreasing the
stipulated time to respond which will reduce the potential damage to the environment .

2. To reduce false alarms of video-based systems, especially in hardly accessible terrain, a

remote controlled UAV can fly to the place where a fire is assumed to confirm that the origin of
the smoke is most likely a fire. The UAV is equipped with semiconductor gas sensors because of
high immunity against disturbances like steam, fog, dust pollution and condensing water. Thus, a
good discrimination between alarm and false alarm is possible.

3. In future we would create a UAV for wildfire detection

4. Generate Dataset using GAN for more accurate prediction of model

29