Volume 6, Issue 6, June – 2021 International Journal of Innovative Science and Research Technology

ISSN No:-2456-2165

Detection of Flood Images using Different Classifiers

Roshni R Menon Sandhra Simon, Maria A S, Rosemaria Shaju, Sachin
Asst. Professor UG Students
Dept of Computer Science and Engineering Dept of Computer Science and Engineering
SCET, Thrissur SCET, Thrissur

Abstract:- Flood is not a new disaster that we face We need to know about the flood is happening nearby
nowadays in every part of the world. It is sudden, fast, or should take preventive measures at the earliest. For, this
and the impact is beyond the imagination. Its frequency we are proposing a method known as Detection of Flood
is increasing day by day. Although we can't avoid this using different classifiers. Here we use different satellite
natural disaster, We should manage it properly. For images to detect the occurrence of flood. These images are
that, image detection has a great role and should find the fed to different machine learning algorithms and predicting
best classifier to detect it. The classifiers we use are k- the accuracy of each classifier.
nearest neighbors, Logistic Regression, Support Vector
Classifier, Decision Tree, and Random Forest machine This paper deals with the 2018 Kerala flood. The
learning algorithms. By learning through each algorithm heavy rainfall and bad weather conditions have taken Kerala
we found the best among them. The accuracy obtained in its most worse situation. Every dam was overflowed in
by learning each algorithm on our trained model is quite such a short time. The network had become down fast and
different and we found out the best. First, we prepared communication was not that much easy for the rescue team
the image dataset which includes remote dataset and as well as the people around there. In such cases, rescue
satellite images. Second, we passed the dataset to each teams need to know the current situation in the field. This
classifier and obtained the variant accuracies. Best information can be best mediated if the affected area is
results are produced in each method. The classifier mapped in real-time. Our procedure assures rapid and
which gives the best can be taken for the early prediction accurate mapping using machine learning techniques. The
of the flood. By using new technologies to manage the best dataset have taken for the training the model. Tested
flood will help us with evacuation faster and take care of using different classifiers and better classifier with best
people who are affected. Flood prediction has done here accuracy is taken.
using history rainfall data so that we just predicted the
chance. Detection is done mainly with high accuracy and A. PROBLEM DOMAIN
the accuracy of each classifier is shown. Also the image Flooding detection has a very important role in
tested result shown. managing the flood in the affected area. Existing systems
use fewer data and thereby getting less accuracy for the
Keywords:- Flood Detection ; Accuracy ; Training ; detection. The dataset including satellite images is used for
Convolutional Neural Network ;Logistic Regression ; K- detection. Moreover, the use of images taken from remote
Nearest Neighbor ; Naive Bayes Classifier ; Support Vector areas should also consider for the exact detection. Those
Machine ; Synthetic Minority Oversampling Technique. images gave more accurate segments for the image
segmentation. Today there are a variety of techniques that
I. INTRODUCTION we can choose from. The important role of images and
classifiers is crucial in each technology or whatever the
Flood become a frequent disaster that occurs in every techniques are using detection of image using better
part of the world. In India, almost every year in new places classifier is the best choice. The main drawback of the
the floods are born and which is almost uncontrollable. The existing system is not knowing the better classifier. Some of
government is hardly fighting for the management before, the systems use Twitter images combining with Artificial
during, and after occurred. Satellite images have increased Intelligence to detect which is more time-consuming. There
the ability to predict the flood and thereby preventing the is also a technique that uses hardware at each flood-prone
loss of lives, property, and homes. It can help in the early area, detecting the factors of a flood using sensors and
evacuation of nearby places. Since the number of satellites predicting the occurrence of flood and all. The main demerit
and sensors have increased their prediction of a flood. of these things is network gets down in those areas when a
Significant improvements in the near real-time assessments flood occurs, thereby no further information about that area
of floods have been made due to increased data acquisition can be received. In that cases, detecting the flood using
rates, higher sensor resolution, the improvement of change machine learning holds a huge role and thereby finding the
detection algorithms, and the integration of remote sensing best algorithm to train the model. Thus it has to be detected
systems. We cannot predict these floods at their full in a very accurate way. We use machine learning algorithms
accuracy just can make sure of the lives as possible as that give perfect results. The dataset taken should be
earliest. accurate for classifying. Images act as a weapon for
detecting the disaster at the most valuable time. Using recent
floods in the last years can be used for high accuracy in the

IJISRT21JUN524 www.ijisrt.com 761

Volume 6, Issue 6, June – 2021 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
detection. The suggested classification system for flood agriculture field a testing image it will give the result as Not
detection looks promising and addresses the issues raised by Flooded. The confusion matrix representing the prediction
the previous systems. The data is collected in real-time with label is shown in the Figure 1.
high-resolution cameras which eliminates the need for a
poor classifier to detect the images. The suggested classifier
will be better.

B.PROPOSED METHOD
In this current scenario, natural disasters are more
common compared to previous environmental situations.
Flooding due to heavy rainfall is the major issue that we
face in our daily life. Our project is focusing on the early
flood detection that can take place by analyzing different
satellite images. Floodnet dataset is used to train with
different machine learning classifiers. The main four
classifiers that we applying are SVM, KNN, Logistic
Regression, Decision Tree, etc. we also input the testing
images additionally for checking the result of each
algorithm used. It shows the result as this is Flooded or Not
Flooded. After the execution there we also compare the
accuracy obtained by each algorithm and will take the best Fig 1.Confusion Matrix of Flood detection.
classifier leading with a higher accuracy rate.
The models are used for the comparison between the
C. DATASET accuracy of detecting flood. Each classifier will be trained
The dataset that we are implementing for our project is with the given dataset and predict the output. Flooded and
mainly from the Floodnet Dataset. This data set from the No-flooded images are shown in the Fig1.1 & Fig 1.2.
Microsoft teams are basically collected by the International
Conference combined with UMBC for the Satellite image
data learning. This study is utilizing for research on different
satellite images related to weather, natural disasters like
floods, cyclones, and to find different hydrological
parameter ,etc. Here, this dataset is classified as about 500-
1000 flooded and non-flooded images. Besides, this
information is more useful for exploring the climatic
changes happening nowadays. Heavy rainfall is occurring
due to the high rate of precipitation and humidity. Which
leads to the occurrence of the life-threatening disaster
known as Flooding. These satellite images are then trained Fig 1.1 Flooded Images
with different machine learning algorithms.

The algorithm that is used for further training is CNN

(Convolutional Neural Network). The system will try to
learn the features, attributes of the particular image. For
example, if the image containing buildings, short-term
streams, playgrounds, etc may be used to be classified
accordingly. The image is labeled as using the Longitudinal
and Latitude values. The data divided into two processes,
tests and the validation part. We use 60 to 70percentage of
data to train the machine. Balance 30 percentage data is Fig 1.2 Non-Flooded Images
given into the test folder for validation if it works or not.
Here we use the mobile net model which preprocesses the Training network used here is CNN(Convolutional
input on your inputs before passing them to the model. Neural Networks). It is also a Multilayered neural network
Dividing each into three batches given ten as a size. Image representing different classes. CNN models can detect
data generator is the function used. using this function we complex features in data. In this, we used satellite images
converting into a machine-understandable form. Thus for the classification for detecting whether it is flooded or
actually machine studies this and show the output as how Not flooded. CNN is also used for image classification and
many images belong to each class. The library called Keras recognition. The weight of each data is shared between the
is used in the model. And find the above-mentioned inputs. Thus, weight sharing is an advantage in this system.
features and divides them into different classes. After the
learning process, it will be tested using an image whether it
is flooded or not. For example, If we are giving an image of

IJISRT21JUN524 www.ijisrt.com 762

Volume 6, Issue 6, June – 2021 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
II. EVALUATION ALGORITHMS ● YES or NO
● True or False
● KNN(K-Nearest Neighbours) ● High or Low
It is a super simple way of classifying data and
supervised machine learning models. It classifies the new Logistic regression used in our project is used to
data point into the target class, depending on the features of classify different satellite images to detect flood. Variable
its neighboring data points. Features of KNN model, it is a used is (logis_clf = LogisticRegression()).The variable used
classification algorithm and a Non-Parametric, which means for training the elements are (logis_clf.fit(X_train,
it does not assume any particular kind of distribution or y_train))the prediction variable used here is(y_pred =
functional relationship of the data that we are trying to logis_clf.predict(X_test)). The main draw backs of logistic
predict or learn the pattern. It is also used to transform regression is mainly Overfitting.
underlying data the features into a higher dimension after
that parametric approach is applied. ● Decision Tree
A Decision Tree is a tree-shaped diagram used to
The training process includes different variables to determine a course of action. Each branch of the tree
store the elements like function variables to represent each represents a possible decision, occurrence, or reaction.it is a
section like which classifier, accuracy rate, etc. For example graphical representation of all the possible solutions to
(knn_clf) represents the knn classifiers used, the accuracy decisions. These decisions are based on some conditions. It
score is calculated as such using the variable, is mainly classified as the root node, leaf node, splitting,
(metrics.accuracy_score). Training elements used branch or subtree, parent/child node, pruning.
are(knn_clf.fit(X_train, y_train). Then testing the image is
then fed to the algorithm to check whether it is flooding or The implementation of the Decision Tree in our
not. For example, we used the project is consists of different elements to train the particular
(filename='evaluate/flood4.jpg') and the result was it images etc. The decision tree classifier is represented as
belongs to the flooded class. The drawbacks of the KNN dec_clf. Training the model using the training sets
algorithm are outlier sensitivity and missing value (dec_clf.fit(X_train, y_train)).Predicting the response for
treatments. test dataset (y_pred = dec_clf.predict(X_test)).

● SVM (Support Vector Machine) III. SYSTEM ARCHITECTURE

It is a discriminative classifier that is formally
designed by a separative hyperplane. It is a representation of
examples as points in space that are mapped so that the
points of different categories are separated by a gap as wide
as possible. It is also a supervised learning algorithm that
can be used for classification or regression problems. It
works well even with unstructured and semi-structured data
like text images and trees.it is maximum margin hyperplane,
finds the linear model with maximum margin unlike the
linear classifiers, the objective is not minimizing the sum of
squared errors but finding a line/plane that separates two or
more groups with maximum margins.

Support Vector Machine used in this project for the

classification of different satellite images and detection of
flooding process. The classifier used here is the (svm_clf).
Train the model using the training sets using the function
variable (svm_clf.fit(X_train, y_train)). The training
elements used are (X_train, y_train).Predicting the dataset
using the variable using (y_pred = svm_clf.predict(X_test)). IV. WORKING EXPLANATION
The demerits of the SVM it is not suitable for large datasets.
The problem domain that we are facing floods due to
● Logistic Regression unconditional climatic changes and heavy rainfall. So to
Regression analysis typically is represented as the detect the flood we should be alert on upcoming flooding.
predictive modeling technique.it estimates the relationship Thus, We propose the method of flood detection using
between a dependent (target) and an independent image processing with different training machine learning
variable(predictor).It is also a supervised learning algorithm. classifiers. Different types of steps are taking place in this
Logistic Regression produces results in a binary format detection phase of flooding.
which is used to predict the outcome of a categorical A.PSEUDOCODE
dependent variable. So the outcome should be 1. Importing necessary packages
discrete/categorical such as: 2.Collection of Satellite Images
● 0 or 1 3. Flood net Dataset preparation.

IJISRT21JUN524 www.ijisrt.com 763

Volume 6, Issue 6, June – 2021 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
4. Image Classification labeled as Flooded & Non -
Flooded.
5. Implementation of Different ML Classifiers.
6. Different training method algorithms like KNN, SVM,
Logistic Regression, Decision Tree
7. Output is represented as Flooded or Non-Flooded
8. Higher Accuracy and Precision are found out on each
Classifier.
9. Comparing Higher Accuracy Classifiers is done.

V. HARDWARE SPECIFICATIONS

● Processor : i5
● RAM : 4 GB VIII. CONCLUSION
● Hard Disk : 500 GB
Flood Detection using different algorithms was
VI. SOFTWARE SPECIFICATIONS successfully classified using well balanced dataset. For
balancing dataset we used SMOT analysis in which used for
● Python accurate classification. For better classification the use of a
Python is most simplest language that can be used for dataset and the most accurate classifier is essential.
various tasks such as classification, Software Development Mobilenet data model is used for converting the data into
and even many more. Python 3.7 is used here. batches with 10 as the size for each. To compare accuracies
we used the four best among them and from that, SVM is
● IDE : SPYDER the best to detect the flood dataset. This is the optimal
For the programming in python language we use open classification algorithm in which we tested for binary
source cross platform called Spyder here.Spyder is best used classification labeled as flooded or no-flooded.
with jupyter notebooks or other scientific computing tool
such as Anaconda which provides a general python Among the classifiers we used Logistic Regression,
development environment. Support Vector Machine crossed the real accuracy level that
exists. After all, more input data and machine learning
● Scikit-learn (Sklearn) algorithms are needed to be tested if we wish to find the best
Package used for accuracy calculation,modelling in combination for detecting flooded areas. The accuracy has
machine learning which helps us also to draw confusion to be increased shortly for better real-time flood
matrix. management.

● Keras ACKNOWLEDGMENT
The library used for python interfacing for neural
networks. We offer our heartfelt gratitude to everyone ho helped
us to make this effort a grant success. We thank the
VII. PERFORMANCE EVALUATION almighty God for all the benefits he has bestowed upon us.
Our guide, Prof. Roshni R Menon deserves a lot. Our
Classifier Algorithms gave different accuracy levels principal, DR. Nixon Kuruvila, and our Department Head
using the same dataset are shown in the table 1.1. Dr. M Rajeswari, for unwavering support throughout the
project.
Logistic Regression 91.79978700745474 REFERENCES
Support Vector Machine 93.9297124600639. [1]. Lorenzo Alfieri, Sagy Cohen, John Galantowicz, Guy
JP Schumann, Mark A Trigg, Ervin Zsoter, Christel
K-Nearest Neighbour 74.54739084132055 Prud-homme, Andrew Kruczkiewicz, Erin Coughlan
de Perez, Zachary Flamig, et al, (2018) , "A global
Decision Tree 87.43343982960596. network for operational flood risk reduction" .
Table 1.1 Environmental science & policy, 84:149–158.
[2]. Serena Ceola, Francesco Laio, and Alberto Montanari,
The training success with each machine learning (2014) "Satellite nighttime lights reveal increasing
method is presented with the percentage of correctly human exposure to floods worldwide" . Geophysical
classified training points. Research Letters, 41(20):7184–7190,
[3]. Ben DeVries, Chengquan Huang, Megan Lang, John
Jones, Wenli Huang, Irena Creed, and Mark Carroll,
(2017) "Automated Quantification of Surface Water

IJISRT21JUN524 www.ijisrt.com 764

Volume 6, Issue 6, June – 2021 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
Inundation in Wetlands Using Optical Satellite [16]. K. Panchal, H. Ajmani, and S. Y. Sait,(2019)
Imagery“, Remote Sensing, 9(8):807. ‘‘Flooding level classification by gait analysis of
[4]. Laura Giustarini, Renaud Hostache, Patrick Matgen, smartphone sensor data,’’ IEEE Access, vol. 7, pp.
Guy J.P. Schumann, Paul D. Bates, and David C. 181678–181687.
Mason, (2013) "A Change Detection Approach to [17]. K. Tuberson and K. Tremaine, (2017) ‘‘How the
Flood Mapping in Urban Areas Using TerraSAR-X". Internet of Things can prepare cities for natural
IEEE Transactions on Geoscience and Remote disasters,’’ Harvard Bussniss Review.
Sensing, 51(4):2417–2430. [18]. D. B. Soubhagyalaxmi, L. S. Bhosle, T. Swetha, P. D.
[5]. Fabio Cian, Mattia Marconcini, and Pietro Ceccato, Veena, and B. V. Chetan, (2018) ‘‘Disaster
(2018 ) "Normalized Difference Flood Index for rapid management system using IoT,’’ Int. J. Res. Trends
flood mapping: Taking advantage of EO big data". Innov., vol. 3, no. 6, pp. 148–154.
Remote Sensing of Environment, 209:712–730. [19]. X. Chen, X. Zhu, and D. Zhang, (2010) ‘‘A
[6]. Hanqiu Xu. (2006) "Modification of normalised discriminant bispectrum feature for surface
difference water index (NDWI) to enhance open water electromyogram signal classification,’’ Med. Eng.
features in remotely sensed imagery" . International Phys., vol. 32, no. 2, pp. 126–135.
Journal of Remote Sensing, 27(14):3025–3033. [20]. T. K. Ho, (2002)‘‘A data complexity analysis of
[7]. Jian Li and David Roy. (2017) "A Global Analysis of comparative advantages of decision forest
Sentinel 2A, Sentinel-2B and Landsat-8 Data Revisit constructors,’’ Pattern Anal. Appl., vol. 5, no. 2, pp.
Intervals and Implications for Terrestrial Monitoring". 102–112.
Remote Sensing, 9(9):902.
[8]. Igor Klein, Andreas Dietz, Ursula Gessner, Stefan
Dech, and Claudia Kuenzer, (2015). "Results of the
Global WaterPack: a novel product to assess inland
water body dynamics on a daily basis". Remote
Sensing Letters, 6(1):78–87.
[9]. Sandro Martinis, Andre Twele, and Stefan Voigt.
(2009) " To wards operational near real-time flood
detection using a split based automatic thresholding
procedure on high resolution TerraSAR-X data"
Natural Hazards and Earth System Sciences, 9(2):303–
314 ,
[10]. Perry C Oddo and John D Bolten. (2019) "The value
of near real time earth observations for improved flood
disaster response". Frontiers in Environmental
Science, 7:127,
[11]. Jean-Franc o̧ is Pekel, Andrew Cottam, Noel Gorelick,
and Alan S Belward. (2016) " High-resolution
mapping of global surface water and its long-term
changes". Nature 2.
[12]. Stefan Schlaffer, Patrick Matgen, Markus Hollaus, and
Wolf gang Wagner. (2015) "Flood detection from
multi-temporal SAR data using harmonic analysis and
change detection". International journal of Applied
Earth Observation and Geo information, 38:15–24.
[13]. Yuxin Wu and Kaiming He. (2018) "Group
normalization, In Proceedings of the European
Conference on Computer Vision" (ECCV), pages 3–
19, 5.
[14]. A. Al-Fuqaha, M. Guizani, M. Mohammadi, M.
Aledhari, and M. Ayyash, (2015) ‘‘Internet of Things:
A survey on enabling technologies, protocols, and
applications,’’ IEEE Commun. Surveys Tuts., vol. 17,
no. 4, pp. 2347–2376, 4th Quart.
[15]. A. Y. Sun and B. R. Scanlon, (2019). ‘‘How can big
data and machine learning benefit environment and
water management: A survey of methods, applications,
and future directions,’’ Environ. Res. Lett., vol. 14, no.
7, p. 28.

IJISRT21JUN524 www.ijisrt.com 765