A REAL-TIME FOREST FIRE

DETECTION SYSTEM USING

DEEP LEARNING

Submitted By
Mr. MONISH KUMARAN.L Reg No.203031101078
Mr. RAGUNANTHAN.C Reg No.203031101097
Mr. RAJAPANDIYAN.V Reg No.203031101100
ABSTRACT

Predicting the spread of forest fires is essential for preventing destruction. This is a serious environmental
issue that leads to ecological destruction in the form of a landscape threatened by natural resources,

which in turn disrupts the stability of the ecosystem, raises the risk for other natural hazards, and reduces
resources like water, which leads to global warming and water pollution.

The goal of this research is to develop a deep learning model based on computer vision that can monitor and
report forest fires.

The proposed model approach in this paper presents CNN algorithm detection of forest fires with high
accuracy to detect the fire in a particular area with faster and reliable.
INTRODUCTION
On average, wildfires in the United States damage 1.2 million acres of forest each year.

Forest fires in the Asian country increased by a factor of 125 between 2016 and 2018.

For this reason, it is crucial that fire detection technologies and systems be continually
researched, developed, implemented, and improved.Predicting, preventing, and
managing forest fires has grown in significance.

Several fire risk models at present make use of forest fire databases to build and
evaluate probabilistic models.In order to predict when fires are most likely to break out
in a certain devised a model that takes into account the area's elevation above sea
level, the dates of previous fires, and the dates on which no fires broke out.
LITERATURE SURVEY
01 04
Wenkai Yan Yuncong Li
A Deep Learning Method based on SRN-
“A New Forest Fire Risk Rating Forecast YOLO for Forest Fire Detection
Model Based on XGBoost”- 2022
03
Alexander Soderlund
Estimating the Spread of Wildland Fires via Evidence-Based
02 Information Fusion-2023 05
AbdelHamid Nassar Ruixian Fan
A Machine Learning-Based Early Forest Fire Lightweight Forest Fire Detection
Detection System Utilizing Vision and Sensors’ Based on Deep Learning-2021
Fusion Technologies
SYSTEM ANALYSIS
EXISTING SYSTEM

It is true that frequent fires on large scales cause air pollution, mar quality of stream water,
threaten biodiversity and spoil the aesthetics of an area, but fire plays an important role in
forest ecosystem dynamics.

In the existing system based model using data with a similar number of images for each class.

There is no way to detect fire or smoke using the current system, and there is also a high
chance that the warning is false.

DISADVANTAGES

Several categories represented by a set of interchangeable, visually similar images.

Reduced precision
Fire wasn't properly identified.
PROPOSED SYSTEM
In our proposed system propose the deep transfer learning approach with CNN pretrained
models to detect a Fire in forest area.
Future forest fire damage is likely to be mitigated by improved fire prediction methods.
There are many different types of fire detection algorithms available,here we proposed CNN
algorithm.
Based on training time and epoch number, this work presents the overall classification
accuracy rate of three pretrained architectures. At first perform the data collection process
and preprocessing.
Second, we split our dataset into three segments required for training, testing, and validation.
Third perform CNN pretrained based training model for training image. Finally predict the fire
in the forest area from given testing image.

Advantage
Fast and easy to identify.
Reduce the potential damage.
Improve the classification accuracy.
DEEP LEARNING
Deep learning is a subset of machine learning, which is
essentially a neural network with three or more layers.

Allowing it to “learn” from large amounts of data.

Deep learning models are capable enough to focus on the accurate features themselves by
requiring a little guidance from the programmer.
PYTHON
Python is an interpreted, high-level, general-purpose programming language.

Less Code Massive Community Support

Pythons support for pre-defined Python has a huge community of users
packages, we don’t have to code which is always helpful
algorithms

Platform Independent Prebuilt Libraries

Python can run on multiple
platforms including Windows, Jupiter is a gas giant and the
MacOS, Linux, Unix, and so on. biggest planet
Dataset collection
The dataset acquired in this study is a collection of images of forest
fire area and without fire area. There exist around 256 raw images of
different dimensions (width height), usually measured in terms of pixel
values.

The sample Forest images are gathered from the Kaggle dataset. The
collected images are in Joint Photographic Experts Group (JPEG)
format. The image database is categorized into two segments, Fire and
no fire. Normal based on the existence of the forest image.

Generally, in our work, we split our dataset into three segments

required for training, testing, and validation.
Preprocessing:
Due to machinery limitations could be found in forest fire images. The abnormalities such as poor
quality image resolution, distortion, inhomogeneity, misinterpretation, and motion heterogeneity are
produced by limitations in forest image processing.
The CNN-pretrained models require the fire and without fire in forest to be resizedwith a 224 × 224 × 3
dimension , so the dataset forest images are reformatted to a specific dimension.
Transformation:

Data comes in all shapes and sizes: from images to text to time series data. A simple Excel spreadsheet might have data
in a few columns, while a more complex BigQuery dataset could have millions of rows and thousands of columns.
Transfer learning (TL):
It is a technique in deep learning that focuses on taking a pre-trained neural network and storing
knowledge gained while solving one problem and applying it to new different datasets.
In this article, knowledge gained while learning to recognize 1000 different classes in ImageNet could
apply when trying to recognize the things.

Classification: In this module give input image to the Training model. This training model can
predict the type of image such as fire and without fire.
Testing
Unit testing
Unit tests are usually automated and are designed to test specific parts of the code, such as a particular
function or method. Unit testing is done at the lowest level of the software development process

Integration testing
Integration testing is a method of testing how different units or components of a software application
interact with each other.

System testing
This software is tested such that it works fine for the different operating systems. It is covered under the black
box testing technique. In this, we just focus on the required input and output without focusing on internal
working.
BLOCK DIAGRAM USECASE DIAGRAM
Screenshot
Conclusions
■ The main goal of this is to design efficient automatic detection of
fire in forest classification with high accuracy, performance and
low complexity.
■ In the conventional image classification is performed. The
complexity is low. But the computation time is high meanwhile
accuracy is low.
■ Further to improve the accuracy and to reduce the computation
time, a convolution neural network-based classification is
introduced in the proposed scheme.
■ Also, the classification results are given as fire and no fire images..
Also, python language is used for implemented for classification.
■ It is one of the pre-trained models. So the training is performed for
only final layer.
REFERENCE
● Alonso-Betanzos, O. Fontenla-Romero, B. Guijarro-Berdi˜nas, E. Hern´andez-Pereira, M. Inmaculada Paz Andrade,
E. Jim´enez, J. Luis Legido Soto, and T. Carballas, “An intelligent system for forest fire risk prediction and fire
fighting management in Galicia,” Expert systems with applications, vol. 25, no. 4, pp. 545–554, 2003.
● N. Aronszajn, Introduction to the theory of Hilbert spaces. Stillwater, Oklahoma: Reasearch [sic] Foundation, 1950.
● D. R. Brillinger, H. K. Preisler, and J. W. Benoit, “Risk assessment: A forest fire example,” in Statistics and Science:
a Festschrift for Terry Speed, ser. IMS Lecture Notes Monograph Series, D. Goldstein and T. Speed, Eds.
Beachwood, OH: Institute of Mathematical Statistics, 2003, vol. 40, pp. 177–196.
● T. Cheng and J. Wang, “Applications of spatio-temporal data mining and knowledge for forest fire,” in ISPRS
Technical Commission VII Mid Term Symposium, Enschede, 2006, pp. 148–153.
● K. Clarke, J. Brass, and P. Riggan, “A cellular automaton model of wildfire propagation and extinction,”
Photogrammetric Engineering and Remote Sensing, vol. 60, no. 11, pp. 1355–1367, 1994.
● P. Cortez and A. Morais, “A data mining approach to predict forest fires using meteorological data,” in Proceedings
of the 13th Portugese Conference on Artificial Intelligence, 2007, pp. 512–523.
THANK YOU