Develop a Robust System for Detecting Counterfeit Iraq Currencies Based on Deep Learning Techniques

Abstract

Counterfeiting poses a serious threat to the financial economy because advanced counterfeit banknotes, thanks to advances in printing technology, have become difficult to identify through traditional investigation techniques. Maintaining the security of international economic systems and keeping pace with evolving counterfeiting strategies requires continuous study and development in this field. The Central Bank of Iraq is the exclusive management responsible for issuing local currency, so verifying the authenticity of Iraqi currency is of utmost importance to maintain the integrity of the country's financial economy. This thesis aims to develop a reliable system for detecting counterfeit Iraqi currency that can distinguish the subtle differences between real and counterfeit Iraqi currency. This work utilizes machine learning algorithms such as Random Forest, XGBoost, Decision Tree Classifier, Support Vector Machine (SVM), and CatBoost.  In addition, deep learning models such as Convolutional Neural Networks (VGG16, InceptionV3, MobileNetV2) were employed to allow counterfeit banknote detection with high accuracy and reliability.The proposed system was trained on a dataset of 1359 images that include two types of Iraqi currencies (real and counterfeit with the highest level of professionalism) and in different categories, as they were collected in cooperation with the Central Bank of Iraq after obtaining official approvals from the relevant authority. The dataset underwent initial processing using augmentation and annotation techniques to increase the dataset number to improve the network's performance in the training process concerning prevalent feature extraction and thus achieve high detection accuracy, becoming 4188 for real currencies and 3966 for counterfeit currencies. The dataset was divided into 80% for training and 20% for validation. In this work, a real-time system was built and implemented based on a set of main components including Raspberry Pi5, Raspberry Pi camera, servo motor, and LCD screen. The device discovers the Iraqi currency using a camera and a servo motor supported by UV light to capture the currency image to ensure the highest clarity and accuracy. The image is sent to pre-trained deep learning and machine learning models to classify it as counterfeit or real. Finally, the detection result is displayed on an LCD screen. The experimental results in which CatBoost and SVM were used showed an accuracy of up to 98%, while the accuracy of the CNN model ranged to 99%. These results demonstrated the effectiveness of advanced technical solutions in thwarting the risks posed by counterfeit money, protecting the financial economy, and reducing losses.