Currency Detection using TensorFlow

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 09 Issue: 04 | Apr 2022 www.irjet.net p-ISSN: 2395-0072
© 2022, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page 2962
Currency Detection using TensorFlow
Shelar Rutuja1, More Smruti1, Tapase Nisha1, Sanjay Waykar1
1MGM College of Engineering and Technology, Kamothe
---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract:Advancesintechnologyhavereplacedpeoplein
almost every field with machines. Thanks to the introduction
of machines, banking automation has reduced the burden on
humans. Banking automation requires more attention to
declining currencyhandling.Whenthebanknoteisblurredor
defaced, it is difficult to identify its currency value. A
sophisticated designisincludedtoincreasethesecurityofthe
call. This makes the call recognition task very difficult. For
correctcurrency recognition, it is very important to choose a
good function andan appropriate algorithm. One of the main
problems that blind people face is the recognition of money,
especially cash. In a way, the seemingly weakened people do
not think about cash settlement and run into problems
related to cash transactions in their daily life. It is a useful
treatment for those who are externally weakened. studies
and trialswere conducted according to key points, such as
watermarks, images printed on money, the value of words
and numbers, and the total amount of information gathering
that stimulated CNN .This paper focuses on the study of
solvingsocial problemsusingConvolutionalNeuralNetworks
(CNNs) and validating and evaluating different CNN models.
Here, the Alexnet, Googlenet and Vgg16 models were
considered for the study .All models were adjusted during
preparation and testing of individual data sets. Among these
three models, Alexnet had the best performance, Vgg16
model showed 100% performance, and Google net showed
performance with 88%.
Keywords- banknote recognition; convolutional neural
networks; computer vision; deep learning;VGG16; transfer
learning.
I. INTRODUCTION
Currency is notes and coins issued by the governmentfor
circulation in the economy. Service and product exchange
facility. Banknotes are an important medium for
trading. The banknote is characterized by simplicity,
durability, full control and affordability. This made him well
known. Compared to all other alternative forms of currency,
paper is the most preferred form of currency. This has one
downside to banknotes: it cannot be reused, but the issue is
less serious compared to other methods. As part of the
technological advancements introduced to the financial
sector, financial institutions and banks have embarked on
financial self-service. An automated banking system is
implemented that processes currencies using, machines
with ATM counters and coin dispensers. In this situation, the
usesa currency recognizer to classify the banknotes. Call has
two types of features i.e. internal features and external
features. External characteristics include physical aspects of
the currency, such as widthand size.However,thesephysical
features are unreliable as circulation can damage the
currency.Thiscompromisedcall systemcancausethesystem
to not recognize the call. Internal features include unreliable
color features because calls go through different hands and
this can get dirty and give incorrect results. For currenciesof
eachdenomination there is a specific color and size that is
followed by Reserve bank of India. It is a very simple for
human to identify them denomination of currency note
because our human brainisextremelyskillfulinlearning new
matters and discovering them later without much trouble.
But this currency recognition task turns very difficult and
challenging in computer vision, in cases when currencies
becomes damaged, old, and faded due to wear and tear of
currencies. Security features are included in every Indian
Currency which dispenses help in recognition and
identification of the currencyvalue.VariousSecurityfeatures
are identification mark (shape), Center value,Ashoka,Latent
image, See through register, Security thread, Micro letter,
Watermark and RBI seal.
1.1 Deep Learning: Deep Learning is a new machine
learning environment, introduced with the aim of bringing
Machine Learning closer to one of its original motive:
Artificial Intelligence ”(p.6). In-depth reading refers to the
study of multiple categories of illustrations and concepts
that assist researchers in analyzing data, including images,
sounds, and texts
In-depth reading is often associated with a neural
network witha few layers that can learn from alargeamount
of data, which includesa series of labeled images. In addition
it has been widely used in the field of vision and voice
(processing natural language). The weights of any layer are
learned through the spread of back propagation in an in-
depth learning process. All layers have different effects on
data analysis. Despite its difficulties, the method has been
successfully extended to a variety of classification and
identification problems.
1.2 Deeplearning techniques:
Classic-Neural-Networks: A multilayer perceptron,
inwhich neurons are associated with a continuous layer,and
is often used to identify Neural Connected Networks. There
are 3 functions involved in the launch system:Linefunction:
Properly called, this indicates only the line that repeats
the input with a continuous repetition. Non-line work:
This is categorized into three categories: A- “Sigmoid Curve:
A function that is translated as anS- shaped curve with its
width starting from 0 to 1.

B- Hyperbolic tangent (tanh) refers to an S-shaped
curve with a distance of 1 to 1.
C-Rectified Linear Unit (ReLU): A single point function
that returns 0 when the inputvalue isbelowthesetvalueand
the line multiplication if the input value is high.
Convolutional-Neural-Networks: CNN is a continuous
version of the highly developed neural- network-model
model. It is designed to handle high levels of complexity, as
well as pre-processing and data integration.
CNN is divided into four stages after the input data was
presented in the conversionprogram:Convolution:aprocess
that generates feature maps from input databases and uses
function in these maps.
Max Pooling: Assists C-NN in image acquisition based
on imaginary modification.
Plans: At this stage, the data that has been generated
is thenpressed for CNN to check.
Complete Connection: This is sometimes described as
a hidden layer that collects system
loss function.
Recurrent-Neural-Networks(RNNs):initially,RNN was
developed to assist insequencingprediction;forexample,the
Long-Short-Term-Memory (LSTM) algorithm is widely
knownfor its versatility. These networks are based solely on
datatracking with variable input lengths.
Generative-Adversarial-Networks: Using an efficient
production system for counterfeit production models and
detecting real-timedata fromproducers,knownas(GANs).In
their research, they used GANs and could distinguish
between fake and real banknotes with a high degree of
accuracy. GAN.s is a two-dimensional and highly complex
neural network. The generative- network is the first, and the
racist network is the second. After using GANs in the
database and using them to separate real and fake notes,
promising results were obtained.
1. Single-shot-multibox-detector: It is a comprehensive
mathematical model in finding objects. It creates connecting
boxes using working maps from a variety of layers. It will
builddifferent boundary boxesbasedondifferentcategories;
after that, one can find out what the object was. This system
can also be used for real-time detection, and has faster
capabilities than R_C-NN and Res Net.
2. Multilayer perception: This is one of a kind of neural-
network built into a layer of input and a node that works
togetherto produce one or more hidden layers and effects.
To calculate output, indirect activity is required, and a back-
propagation- algorithm can be used to train MLP isolation
and reduction. In addition, MLPs were a feed forwardneural
network composed of a number of nodes connected by a
single connection and trained using back-propagation.
II.EXISTING SYSTEM
The authors of [1] proposed a method of recognizing money
using monetary features such as central number, latent
image, RBI, form, and micro letter. It contains a trainingdata
set for preparing the training model. PCA analysis accounts
for banknote recognition. In [2], the authors proposed a
method for recognizing banknotes using features such as
color, texture, and size. Dirty banknote detection method
activated. Author [3] proposed a banknote recognition
system using MATLAB tools. PCA analysis according to with
a description of Euclidean distance. Describes the LBP
method for matching purposes In [4], the authors first
determine the country and then propose a method for
determining the country name. The author of[5]gavea brief
introduction to the features of the Indian currency. Shape
filtering process according to with other analysis and
segmentationdescriptions.A banknotedetectionrecognition
technology is implemented using a neural network. [6] is a
book on CNNs which includes various processing of layers
with mathematical demonstrations. Once an image is
captured, it is transformed to matrix form prior to detection
and then matched with a trained model to demonstrate all
processing betweenmatrixgenerationand recognitionusing
mathematical calculation. Effect of various convolution
kernels on horizontal and vertical edges. In [7], the authors
described all aspects of CNNs. In [8], the author proposed a
method for recognizing currency using a neural network.
Classification using weighted
Euclidean distance with various steps required for
data collection and processing. In [9], [10], the authors
proposed a call recognition technique, and is a learning
machine used to train data and obtain accuracy.

INPUT
DATA
DEEP
LEARNING
TEST & TRAIN
CLASSIFICATION
ALGORITHM
CNN
VGG 16
III. METHODOLOGY
Figure 1: Methodology of the system
In the coding section, we will use the in-depth reading of
Keras library on python to build our CNN (Convolutional
Neural Network). We have to include TensorFlow and
Theano working at the end of the Cameras. TensorFlow isan
organization. An open source software for data flow system
across the entire job list. It is a symbolic mathematical
library, and it is and is used in machine learning applications
such as neural networks. Theano is aPython library and is
doing well mathematician to convertand evaluate statistics
expressions, especially those with amatrix value. In Theano,
the figures are expressed using a NumPy-square syntax was
also compiled for optimal performance it can be CPU or GPU
architecture. After installing the required libraries, we train
our model as discussed above. After training and model
testing, weset the amount oftimethatincreasestheaccuracy
of AFCRS in addition to increasingthe number of periods
Data pre-processing- An easy way to get data without
overloading once under the pre-processing of the data set.
The main purpose after pre-processing the data that adding
value base value which is a set of data generated. Main the
advantage ofpre-processing data is that it is better training-
set. For these purposes, we use the Keras library to pre-
process images. The VGG-16 model we use is ours
experiments require input image format 64x 64 x 3, where 3
refers to R, G, B (red, green, blue) parts of the colored picture
and the image should be 64 x 64 pixels in size. We then use
the following three types of pre-image processing in the
original databases, too select our first filter that starts at 64.
a. Image Rescaling - We need to resize the image to
make the model data into a common format for training to
be developed, accurate, and fast. Sine re-scaling feature on
the cameras. To usethis feature, we need to import a certain
library from keraspre-processingas"ImageDataGenerator".
If the re-rating feature is missing or 0, no resume is used,
otherwise we duplicate the data about it given number. This
is done after installing all the others changes. In our model,
we use re-scaling feature like:
rescale = 1. / 255 in both training and testing data sets.
b. Image Shearing - We need to cut the image to make
the training data better, more accurate. We also have shear
range factor in the cameras. This will be incorporated into
the camera'spre-processing library. In our model, we use a
clipboard asshear_range = 0.2. Shear range is usually the
Shear angle in the middle anti-clockwise direction by
degrees, also known as the shear intensity.
c. Perspective Transformations: Applied perspective
transformations on training data to zoom in the range of
zoom_range=0.2 to get the accurate results by learning inan
accurate manner. Zoom range isa floatorlower,upperrange
forrandom zoom. This is also done by importing a library
from keras pre-processing.
2. Training the CNN: Here, after choosing VGGNet for
our model, we optimized VGGNet [4], which is a pre-trained
network. This speeds up the trainingprocess,astheyarefew
and far between really training layers. Training the neural
network, of course it is actually better to start with a
malfunctioning neural network and expose the neural
network with high accuracy. As terms of job loss, we want
our job loss to be the same very low atthe end of training.
This shows that
The neural network has a high level of learning and
accuracy. The problem of network training equals
productivity as loss function with a small error rate. It is also

important effective evenreduces losses because, it turns out
loss is an easy task to improve. Even though there are many
algorithms that make it work and for better performance,we
select ReLU (linear Rectifier unit) as our activationfunction,
and we can select 'adam' as our development work because,
it develops neural networks with faster training. And also
computational
The ReLU step is simple. To see a note of a given currency
as fake or real, we should see the accuracy of VGG-16 a well-
configured model with a set of generated data. It was about
55% on the corresponding test set, though ours the data set
was very small and limited to 200 images, the result is still
very encouraging. When we enlarge our image data set with
real-world samples that can make the model even higher
accurately trained which may result in our results being
higher 80% accuracy is a good indicator of results. Without
full points, the result may be considered to invest heavily in
the database, a was the training and test sets very similar.
Therefore, after careful consideration, we must evaluate the
loss and accuracy of our model, by changing the size of the
collection and epoch. Basically there are 2 cases in deep
learning about loss and accuracy prices.
Figure 2. Testing procedure of the model
IV. RESULTS
The three-layer CNN model used to classify monetary
notes based on systems has yielded 98.50% accurate results
and 15 times. The training database is divided byan 8: 2
rating foroppositevalidation. Multi-measurementtemplates
are used to determine securityfeatures ina currencynote.To
test the model, a scanned image of the currency note will be
uploaded viatheJupyterFileUploadSystemanditsvariability
is predicted by model. The use of CNN has several
advantages, which include, for example, CNN is well known
for its architecture, and the best part is that no feature is
required. The great advantageof C-NN over itspredecessoris
that it can identify an important factor without the need to
interact with people.
Some of CNN's obstacles are,Imagesindifferentpositions
are separated; due to operations such as max pool, the
Convolutional neural network is very slow, If CNN has
multiple layers, the training phase can take longer, if the
machine does not have a powerful GPU and CNN requires a
large Data-set to process trained neural network
Figure 3. Accuracy of CNN
V. CONCLUSION
Today, technology is advancing on a massive scale. The
proposed system can be extended to detect coins as well as
detect counterfeit currency. You can add the names of
countries other than India, you can also compare between
them. It does not give 100% accuracy when image is loaded
externally into practicefolder. You can solve this problem by
optimizing the system. Therefore, the various methods
proposed in this article were successfully implemented and
validated by running experiments on the model. CNN turns
out to be the best feature to do this technique using tflite and
flutter_tts modules. 95% accuracy is achieved with model
classifications. Also, detection of coins works well in this
method.

REFERENCES
[1]Vishnu R, Bini Omman, "Principal Features for Indian
Currency Recognition", 2014 Annual IEEE India Conference
(INDICON), Pune,
India.doi:10.1109/INDICON.2014.7030679
[2]H. Hassanpour, A.Yaseri, G.Ardeshiri,"Feature Extraction
for Paper currency Recognition", 2007 IEEE 9th
international Symposium on signal processing and its
Applications, doi:10.1109/isspa.2007.4555366
[3]Kalpana Gautam, “Indian CurrencyDetectionUsingImage
Recognition Technique",2020IEEEInternational Conference
on Computer Science, Engineering and
Applications(ICCSEA),doi:10.1109/ICCSEA49143.2020.9132
955
[4]Vedasamhitha Abburu, Saumya Gupta, S. R. Rimitha,
Manjunath Mulimani, Shashidhar G. Koolagudi, “Currency
Recognition System Using Image Processing",2017 IEEE
Tenth International Conference on Contemporary
Computing, doi:10.1109/IC3.2017.8284300
[5]Ms. Rumi Ghosh, Mr. Rakesh Khare, “A Study on Diverse
Recognition Techniques for Indian Currency Note",
2013(JUNE) INTERNATIONAL JOURNAL OF ENGINEERING
SCIENCES & RESEARCH TECHNOLOGY, ISSN:2277-965
[6]Jianxin Wu, “Introduction to Convolutional Neural
Networks”, May 1, 2017.
[7]Simon Haykin, "Neural Networks: A comprehensive
foundation", 2nd Edition, Prentice Hall, 1998
[8]Muhammad Sarfraza, “An intelligent paper currency
recognition system", 2015, International Conference on
Communication, Management and Information Technology
(ICCMIT 2015).
[9]Faisal Saeed, Tawfik Al-Hadhrami, Fathey Mohammed,
Errais Mohammed,” Advances on Smart and Soft
Computing". 2021, Springer.
[10]Minal Gour , Kunal Gajbhiye, Bhagyashree Kumbhare,
and M. M. Sharm, “Paper Currency Recognition SystemUsing
Characteristics Extraction and Negatively Correlated NN
Ensemble", 2011, Advanced Materials Research Vols 403-
408 (2012) pp 915-919 © (2012) Trans Tech Publications,
Switzerland, doi: 10.4028.
[11]TeachableMachine,“https://teachablemachine.with
google.com/train/image".
[12]Rubeena Mirza1,Vinti Nanda2 “Design and
implementation of Indian Paper currency authentication
system based on Feature extraction by Edge based
Segmentation using sobel operator” International journal of
engineering Research and development e- ISSN:2278-
067X,p-ISSN:2278-800X, www.ijerd.com Volume 3,Issue
2(august 2012 ) PP. 41-46
[13]Rubeena Mirza, Vinti Nanda, “paper currency
verification System Based On Characteristics Extraction
Using Image Processing”, International Journal of
engineering and advanced technology (iJEAT) ISSN 2249-
8958, Volume-1, Issue-3,February 2012.
[14]Kalyan Kumar Debnath, bSultan Uddin Ahmed,aMd.
Shahjahan “A Paper currency Recognition System using
negatively correlated Neural network Ensemble”, JOURNAL
OF MULTIMEDIA, VOL. 5, NO.6, DECEMBER 2010@2010
[15]Megha Thakur, Amrit Kaur, “VARIOUS COUNTERFEIT
CURRENCY DETECTION AND CLASSIFICATION
TECHNIQUES” International Journal For Technological
Research In Engineering Volume 1, Issue 11, pp.1309-1313,
July-2014

Currency Detection using TensorFlow

More Related Content

What's hot (20)

Similar to Currency Detection using TensorFlow (20)

More from IRJET Journal (20)

Recently uploaded (20)

Currency Detection using TensorFlow