Fake Currency Detection System Using Deep Learning and Advanced

Software Integration

Devika, Janani, Saravana Kumar and Harsitha

Department of Computer Science and Engineering, SRM Institute of Science and Technology, Ramapuram, Chennai, Tamil

Nadu, India

Keywords: Image Classification, Counterfeit Detection, Financial Security, Basic, Convolutional Neural Networks,

Indian Fake Currency Detection.

Abstract: Identification of False Currency System using Deep Learning and Advanced Software Integration is designed

to tackle the escalating difficulty of counterfeit money, which has far-reaching economic consequences. This

study accurately distinguishes between actual and fake currency notes by utilizing deep learning techniques

and digital processing of images. The technology is able to automatically recognize fake currency with high

accuracy by using image analysis techniques and deep learning models taught on enormous amounts of

currency photos precision. The project also integrates advanced software tools for real-time currency

detection, offering a scalable, user-friendly solution for businesses, ATMs, and financial institutions. The key

innovation lies in utilizing Convolutional Neural Networks (CNNs) in conjunction with picture classification

and cloud-based deployment for easy availability and scalability.

1 INTRODUCTION

Counterfeit currency is a growing concern

worldwide, posing serious threats to economies,

businesses, and individuals. Conventional techniques

for identifying counterfeit money, such hand

examination and conventional scanning techniques,

often fall short due to the increasing sophistication

of counterfeiters. To address this challenge, we

propose an advanced Fake Currency Detection

System that leverages deep learning and cutting-edge

software integration for high accuracy and efficiency.

Our system utilizes neural networks using

convolutions (CNNs) and other machine learning

techniques to analyze intricate details of currency

notes, distinguishing genuine from counterfeit with

remarkable precision. By integrating deep learning

with advanced software solutions, the system can

process images from various sources, including

smartphone cameras and scanners, making it

accessible and scalable for banks, businesses, and the

general public.

This paper explores the architecture,

methodologies, and performance evaluation of our

proposed system. We highlight the advantages of AI-

driven detection over traditional approaches, discuss

challenges in implementation, and provide insights

into future improvements. Our research main aim to

support the creation of a strong and trustworthy

counterfeit detection system, stimulating trust in

monetary transactions and financial security. Our

approach combines deep learning with sophisticated

software frameworks to provide a scalable, cost-

effective, and user-friendly solution for financial

institutions, businesses, and the general public. The

system is designed to minimize human intervention,

reduce error rates, and enhance security measures in

cash transactions. This paper delves into the

architecture, training methodologies, and evaluation

metrics of our proposed model, demonstrating its

potential to significantly improve counterfeit

currency detection. Furthermore, we discuss

challenges related to dataset collection, model

training, and real- world implementation while

exploring potential future enhancements to make the

system even more robust.

2 LITERATURE REVIEW

Early counterfeit detection systems relied on manual

inspection, ultraviolet (UV) scanning, and magnetic

354

Devika, , Janani, , Kumar, S. and Harsitha,

Fake Currency Detection System Using Deep Learning and Advanced Software Integration.

DOI: 10.5220/0013912900004919

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 1st International Conference on Research and Development in Information, Communication, and Computing Technologies (ICRDICCT‘25 2025) - Volume 4, pages

354-360

ISBN: 978-989-758-777-1

ink detection. These methods, although widely used,

were prone to mistakes made by people and failed to

detect high- quality counterfeit notes that closely

resembled genuine currency (Gupta et al., 2018).

There has been application of machine learning

methods to automate counterfeit detection.

Algorithms such as Algorithms like Support Vector

Machines (SVM), k-Nearest Neighbors, and Forest

have been applied to classify genuine and fake

currency based on characteristics that have been taken

out. While these approaches improved detection

accuracy, they lacked robustness against complex

counterfeit patterns (Kumar & Sharma, 2020).

Superior performance in picture classification

tasks has been shown by deep learning, namely

Convolutional Neural Networks (CNNs), including

tasks, including counterfeit currency detection. CNNs

effectively extract intricate features such as texture,

microprinting, and security markings, enabling high-

precision classification (Patel & Desai, 2021).

Deep learning models that have already been

trained, like VGG16, ResNet, and MobileNet, have

been leveraged through transfer learning to enhance

detection performance. Studies indicate that transfer

learning significantly reduces training time while

maintaining high accuracy, even with limited datasets

(Chen et al., 2021).

Several studies have explored hybrid models

combining deep learning utilizing image processing

methods like edge detection, histogram equalization,

and key point extraction. These techniques enhance

the ability of models to detect fine details in currency

notes, improving classification accuracy (Rahman &

Hossain, 2019).

A significant challenge in deep learning-based

counterfeit detection is the scarcity of high-quality

counterfeit currency datasets. Recent research has

proposed using Generative Adversarial Networks

(GANs) to synthesize realistic counterfeit images,

thereby improving model generalization (Zhang & Li,

2022).

The proliferation of smartphones has enabled the

development of mobile applications for real-time

counterfeit detection. These applications utilize deep

learning models to process currency images captured

by mobile cameras, providing instant authentication

(Singh & Verma, 2021).

A major limitation of deep learning-based

detection is the fact that AI models are opaque. In

order to shed light on model decision making,

explainable AI (XAI) approaches like Grad-CAM

and SHAP have been investigated. increasing

transparency and trustworthiness (Zhou et al., 2021).

Despite significant advancements, challenges

remain in real world deployment, including variations

in lighting conditions, currency wear and tear, and

dataset limitations. Future research aims to enhance

model robustness through advanced neural networks,

improved data augmentation techniques, and edge

computing solutions (Brown et al., 2023).

3 PROPOSED METHODOLOGY

The Fake Currency Detection System proposed in this

study integrates deep learning techniques with

advanced software solutions to improve real-time

detection capabilities, efficiency, and accuracy. The

methodology consists of multiple stages, including

gathering information, preprocessing, training

models, software integration, and deployment. The

system is designed to operate on multiple platforms,

including mobile devices, scanners, and cloud-based

applications, making it scalable and accessible.

Gathering and Preparing Data: Training a

successful counterfeit detection model requires a

solid dataset. The dataset consists of high resolution

images of both genuine and counterfeit currency

notes collected from various sources, including

financial institutions, public databases, and synthetic

data generated using Generative Adversarial

Networks (GANs)

(Vivek Sharan, 2019).

Deep Learning Model Selection and Training: To

accurately Determine which are authentic and fake.

currency, we employ the model known as a

Convolutional Neural Network (CNN)for its ability

to learn spatial hierarchies in images.

Software Integration and Real-Time Processing:

An integrated software system uses the deep learning

model that has been trained for real-time counterfeit

detection. This system supports multiple input

sources, including scanners, mobile cameras, and

cloud-based services (Yadav ET AL., 2021).

Edge and Cloud Computing Integration:

Deploying lightweight models optimized using

TensorFlow Lite or Open VINO for Realtime

inference on mobile devices.

Performance Evaluation and System Validation:

Comparison with traditional detection methods to

measure improvement in accuracy and efficiency.

Testing on different currencies to ensure

generalization across various denominations and

designs. Latency Analysis to assess the time taken for

detection in real-time applications (K. B. Zende ET

AL., 2020).

Future Enhancements and Security

Considerations: Implementing techniques such as

Grad-CAM to visualize model decision-making.

Fake Currency Detection System Using Deep Learning and Advanced Software Integration

355

1. Blockchain Integration: Securing verified

currency records on a blockchain ledger for

transparency.

2. Continuous Model Improvement: Periodic

retraining with updated datasets to improve

detection accuracy over time.

This proposed methodology leverages deep learning,

image processing, and advanced software integration

to build a scalable, real-time Fake Currency Detection

System. By combining edge computing, cloud-based

processing, and transfer learning, the system offers an

effective and convenient solution. for detecting

counterfeit notes across different platforms and

environments.

4 IMPLEMENTATIONS

4.1 Detecting Counterfeit Cash with

SIMPLE NN

Figure 1: Simple NN architecture.

Data Collection and Preparation: From publicly

available datasets, financial institutions, and synthetic

data generating (GANs), assemble high-resolution

pictures of real and fake money. Capture images

under varied lighting conditions, angles, and

resolutions to improve model robustness. Label

images appropriately to ensure supervised learning

during model training (Vivek Sharan, 2019).

Data Preprocessing: Convert images to grayscale

to reduce computational complexity. Apply

histogram equalization for contrast enhancement. Use

Gaussian filtering and edge detection (Sobel/Canny

filters) to highlight security features.

Model Selection and Training): Select a deep

learning architecture, primarily for the purpose of

classifying and extracting features, Convolutional

neural networks (CNNs). Experiment with pretrained

models (VGG16, ResNet50, inception Net) using

transfer learning for improved accuracy. Utilizing the

model is trained using the Adam optimizer and the

categorical cross-entropy loss function (Yadav et al.,

2021).

Choose an optimization algorithm (like Adam),

the mini- batch size (the number of samples used in

each iteration), the number of epochs (the number of

times the complete dataset is cycled through the

network during training), and the validation data (a

subset of training data used for validation during

training) are some examples of the parameters that are

chosen when choosing training options (training

parameters).

Software Integration and Deployment: Develop a

real-time detection software using Python and

TensorFlow/Py Torch. Deploy the model using

TensorFlow Serving or ONNX Runtime for efficient

inference. Create a REST API using Flask or Fast API

for seamless integration with web and mobile

applications. Design a user-friendly GUI using T

knitter (for desktop) or React Native (for mobile) for

easy interaction. Figure 1 shows the simple NN

architecture.

Real-Time Processing and Detection: Allow users

to capture currency images through a mobile camera,

scanner, or webcam. Perform image preprocessing and

feature extraction before passing the image to the

trained model. Display detection results with

confidence scores, indicating whether the note is

genuine or counterfeit. Provide visual explanations

using Grad-CAM to highlight key features

influencing model decisions (K. Bhushanm et al.,

2024).

Edge and Cloud Computing Integration:

Implement Edge AI models using TensorFlow Lite or

Open VINO for real-time detection on mobile

devices. Enable cloud-based processing via Google

Cloud, AWS, or Azure, allowing businesses to verify

currency remotely. Ensure low- latency detection by

optimizing the model for light weight deployment.

Security and Blockchain Integration (Future

Enhancements): Implement blockchain technology to

store authenticated currency records for enhanced

security. Develop a self-learning AI model that

ICRDICCT‘25 2025 - INTERNATIONAL CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION,

COMMUNICATION, AND COMPUTING TECHNOLOGIES

356

updates with new counterfeit patterns through

periodic retraining(Aakash S Patel, 2019).

Testing and Validation: Perform rigorous real-

world testing on various currencies to ensure system

reliability. Conduct stress testing to analyze system

response under high-load scenarios. Gather user

contribution to improve the system's effectiveness

and usability.

Final Deployment and User Accessibility: Several

metrics are used to assess the model's performance,

such as accuracy (the percentage of correctly predicted

outcomes), F1 score (the harmonic mean of precision and

recall), and categorized images), precision, and recall.

4.2 Fake Currency Detection Using

CNN

Figure 2: Flowchart of the indian currency classification

process using a convolutional neural network.

As shown in Figure 2, the Indian currency

classification process begins with an input image

dataset and concludes with a fake or real currency

prediction.

Dataset Collection and Preprocessing: Gather a

diverse dataset containing both Images of real and

fake currency from multiple sources. Ensure images

are taken under different lighting conditions, angles,

and resolutions to enhance model robustness (Vivek

Sharan, 2019).

CNN Architecture for Fake Currency Detection:

Accepts preprocessed currency images. Extract low-

and high-level currency features (e.g., edges, patterns,

textures). Reduce spatial dimensions while retaining

essential features.

Model Training and Optimization: Use a pre-

trained CNN model (VGG16, ResNet50, or Inception

Net) for improved accuracy through transfer learning.

Train the model using cross-entropy loss and

optimize using the Adam optimizer. Divide the

dataset into three categories: testing (10%), validation

(10%), and training (80%). for performance

evaluation. Monitor training performance based on

criteria such as F1-score, recall, accuracy, and

precision.

Model Evaluation and Performance Metrics: Use

a to examine false positives and false negatives, use

the confusion matrix. Evaluate model performance on

unseen currency images to ensure generalization

across different denominations. Compare CNN

results with traditional image processing methods to

measure improvements (Yadav et al., 2021).

Real-Time Detection and Deployment: Deploy

the trained CNN model using TensorFlow Serving,

ONNX Runtime, or Flask API for real-time inference.

Integrate with mobile and web applications, allowing

users to scan currency using a camera or scanner.

Display detection results with confidence scores,

ensuring transparency in decision- making.

Edge and Cloud Computing for Scalability:

Optimize CNN using TensorFlow Lite or Open VINO

for fast inference on edge devices. Implement cloud-

based processing using Google Cloud, AWS, or

Azure for large-scale counterfeit detection.

Future Enhancements and Security

Considerations: Implement Explainable AI (XAI)

techniques to provide insights into CNN decision-

making. Integrate blockchain technology to store

verified currency records securely. Continuously

update the model by training with new counterfeit

currency data (K. Bhushanm et al., 2024).

5 RESULTS

The results of the project will be as follows: CNN

Fake Currency Detection System Using Deep Learning and Advanced Software Integration

357

Figure 3: Epoch details of CNN.

Figure 4: Confusion matrix of CNN.

Figure 5: Training progress of CNN.

Figure 6: Epoch details of simple NN.

Figure 7: Confusion matrix of simple NN.

Figure 3,4,5 shows the Epoch, Confusion and

training progress of CNN. And Figure 6,7,8 shows the

Epoch, Confusion and training of the SNN model. As

shown in Table 1, the CNN model achieved perfect

scores in all metrics, whereas the SNN model

obtained slightly lower performance, with 97.7%

accuracy, 0.95 recall, and an F1 score of 0.97.

Figure 8: Training Progress of Simple NN.

Figure 9: Outputs.

ICRDICCT‘25 2025 - INTERNATIONAL CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION,

COMMUNICATION, AND COMPUTING TECHNOLOGIES

358

Table 1: Comparison between both model.

CNN SNN

Accuracy 100% 97.7%

Precision 1 1

Recall 1 0.95

Fl Score 1 0.97

6 CONCLUSIONS

The rapid advancement of counterfeit currency

production has necessitated the development of

intelligent and automated detection systems. This

paper presented a deep learning-based Fake Currency

Detection System, integrating Convolutional Neural

Networks (CNNs) with advanced software solutions

to enhance accuracy, efficiency, and real-time

detection capabilities. Through data preprocessing,

deep learning model training, and software

integration, the system effectively differentiates from

real to fake money notes with extreme accuracy. The

use of transfer learning, cloud computing, and edge

AI ensures that the model Its flexible, scalable, and

able to provide real-time detection on mobile and

desktop platforms. Furthermore, the incorporation

o Explainable AI (XAI) and blockchain technology

offers additional layers o transparency and security.

Despite the system's high accuracy, challenge remain,

such as changes in the illumination, image quality,

and evolving counterfeit techniques. Future

enhancements will focus on improving dataset

diversity, optimizing model efficiency, and

integrating real-time learning mechanisms to

continuously adapt to new counterfeit threats.

7 LIMITATIONS

The following could be the project's limitations:

Dependence on High-Quality Images: The accuracy

of the system heavily relies on clear and high-

resolution images of currency notes. Poor lighting

conditions, motion blur, or low camera quality can

affect detection performance (Vivek Sharan, 2019).

Limited Generalization Across Currencies: The

model is often trained on specific currencies and may

not generalize well to new or less frequently used

banknotes. Differences in currency designs, security

features, and printing techniques may require

retraining for different regions.

Computational Requirements Deep learning

models' inference and training procedures, especially

those based on CNN architectures, require a

substantial amount of processing capacity. Running

high-accuracy models on edge devices (mobile

phones, embedded systems) may require optimization

techniques like TensorFlow Lite or ONNX (Yadav et

al., 2021).

Real-Time Processing Delays: While real-time

detection is a goal, processing large images or

performing deep learning inference on low-power

devices may introduce latency issues. Cloud-based

detection can help, but it depends on internet

connectivity and server response time.

8 FUTURE WORKS

The proposed

Fake Currency

Detection

System

has demonstrated encouraging outcomes in detecting

counterfeit currency using deep learning and

advanced software integration. However, there are a

number of areas that could be improved in the future

its accuracy, efficiency, and scalability. One key focus

is expanding the dataset by incorporating a more

diverse collection of currency notes from different

countries, lighting conditions, and real-world

scenarios. This will improve the model’s

generalization and robustness against variations in

counterfeit techniques. Additionally, Using

Generative Adversarial Networks (GANs) to

generate synthetic data can assist in producing high-

quality fake samples for better training. To further

enhance detection accuracy, future work will explore

more advanced deep learning architectures, such as

Vision Transformers (ViTs) and Efficient Net, which

have shown superior performance in image

classification tasks. Implementing self-learning AI

models that can continuously update and adapt to new

counterfeit strategies will also be a key advancement.

Moreover, optimizing the real-time model

performance by employing techniques like

quantization, model pruning, and TensorFlow Lite

will allow efficient deployment on mobile phones and

embedded systems are examples of edge devices.

REFERENCES

A. Yadav, T. Jain, V. K. Verma and V. Pal, "Evaluation of

Machine Learning Algorithms for the Detection of Fake

Bank Currency," 2021 11th International Conference

on Cloud

Computing,DataScience&Engineering(Confluence),20

21,pp.8108 15,doi:10.1109/Confluence5164 8

.2021.9377127.

Fake Currency Detection System Using Deep Learning and Advanced Software Integration

359

Aakash S Patel, “Indian Paper currency detection”

International Journal for Scientific Research &

Development (IJSRD), Vol. 7, Issue 06, ISSN: 2321-

0613 (June 2019)

Anju Yadav,Tarun Jain,Vivek Kumar Verma,Vipin Pal

“Evaluation of Machine Learning Algorithms for the

Detection of Fake Bank Currency” IEEE [2021]

Archana M Kalpitha C P, Prajwal S K, Pratiksha N,”

Identification of fake notes and denomination

recognition” International Journal for Research in

Applied Science & Engineering Technology

(IJRASET), Volume. 6, Issue V, ISSN: 2321-9653,

(May 2018)

Chakraborty, S., & Pal, S. (2017).* "A deep learning

approach for counterfeit currency detection."

Proceedings of the International Conference on

Computational Intelligence and Networks (CINE),

2017, 89-95.

G.Hariharan , D.Elangovan “Proxy Notes Recognition And

Eradication For Betterment Of The Society” Journal

[2020]

K. B. Zende, B. Kokare, S. Pise and P. S. Togrikar, “Fake

Note Detection System,” IJIRT, Vol. 4, No. 1, pp. 4649,

2017. [15]. F. A. B, P. Mangayarkarasi, Akhilendu, A.

A. S, and M. K, “Fake indian currency note

recognition,” vol. 7,pp. 4766–4770, 2020.

K. Bhushanm, M. Asritha, P. Rafiya Sultana, P. Anil

Kumar, S. Mahesh Babu Publication: International

Journal of Advanced Research in Computer and

Communication Engineering, 2024.

Kiran Kamble,Anuthi Bhansali,Pranali Satalgaonkar,

Shruti Alagundgi, “Counterfeit Currency Detection

using Deep Convolutional Neural Network”, 2019

IEEE Pune Section International Conference

(PuneCon) MIT World Peace University, Pune, India.

Dec 18-20, 2019

P. A. Babu, P. Sridhar and R. R. Vallabhuni, "Fake

Currency Recognition System Using Edge Detection",

2022 Interdisciplinary Research in Technology and

Management (IRTM), 2022, pp. 1-5, doi: 10

Prof Chetan More, Monu Kumar, Rupesh Chandra,

Raushan Singh, “Fake currency Detection using Basic

Python Programming and Web Framework” IRJET

International Research Journal of Engineering and

Technology, Volume: 07 Issue: 04 | Apr 2020 ISSN:

2395- 0056

S. Atchaya, K. Harini, G. Kaviarasi, B. Swathi, “Fake

currency detection using Image processing”,

International Journal of Trend in Research and

Development (IJTRD), ISSN: 2394-9333 (2017)M.

Laavanya, V. Vijayaraghavan. “Real Time Fake

Currency Note Detection using Deep Learning”,

International Journal of Engineering and Advanced

Technology (IJEAT) ISSN: 2249 – 8958, Volume-9

Issue- 1S5, December 2019.

Vivek Sharan, Amandeep Kaur,” Detection of Counterfeit

Indian Currency Note Using Image Processing”

International Journal of Engineering and Advanced

Technology (IJEAT), Volume.09, Issue:01, ISSN:

2249-8958 (October 2019)

Y. Neeraja, B. Divija and M. Nithish Kumar, “Fake

CurrencyDetection Using K-nn Technique,”

IJREITSS,Vol. 09, No. 1, pp. 201-205, 2019

Yadav, R.K., Valecha, P., Paliwal, S. (2021). Counterfeit

Currency Detection Using Supervised Machine

Learning Algorithms. In: Joshi, A., Khosravy, M.,

Gupta, N. (eds)Machine Learning for Predictive

Analysis. Lecture Notes in Networks and Systems, vol

141. Springer, Singapore. https://doi.org/10.1007/978-

981-15-7106-0_17

ICRDICCT‘25 2025 - INTERNATIONAL CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION,

COMMUNICATION, AND COMPUTING TECHNOLOGIES

360