Deep Learning in Banking: Colombian Peso Banknote Detection

Link to Project Article: Deep Learning in Banking: Colombian Peso Banknote Detection

Problem Statement

The challenge of distinguishing between genuine and fake banknotes poses a significant problem for individuals. Many people lack the necessary skills to detect counterfeit banknotes, making them vulnerable to scammers. In this project, we aim to tackle this challenge by utilizing a dataset of original and fake Colombian banknotes specifically created for research purposes.

The Colombian peso (sign: $, with Col$ also being used to distinguish it from other peso- and dollar-denominated currencies) is the official currency of Colombia. (Wikipedia)

Approach

In this project, we employ a deep learning algorithm to predict fake Colombian peso (COP) banknotes. The core library utilized is PyTorch, and transfer learning is employed by leveraging pretrained weights from the ResNet series of models.

About the Dataset

Description: The dataset provided can be used for detecting banknote denominations and counterfeit banknotes.

• The dataset consists of 20,800 images, categorized into 13 classes. This includes 6 classes for original banknotes, 6 classes for counterfeit banknotes, and 1 additional category for the background.
• The dataset incorporates variations in illumination, rotation, and partial views of the banknotes. It is divided into three folders: ds1, ds2, and ds3, each containing 20,800 images.
• Each folder further consists of training, validation, and test sub-folders, containing the respective images.
• The dataset is balanced, ensuring an equal number of images for each class.

Feel free to connect with me on social media:

Twitter: @InuwaAbraham

Analytics Vidhya: Author - Inuwa Mobarak

LinkedIn: Mobarak Inuwa

Personal Website: Mobarak Inuwa