A Proficient Model to Classify Bangladeshi Bank Notes for Automatic Vending Machine Using a Tiny Dataset with One-Shot Learning Siamese Networks

Abstract

Automatic vending machine is a necessity at this technological era. It's a step to go toward the vendor-less shop management, which supports the commandment of the 4th Industrial Revolution. But image recognition system needs a lot of data to get the pattern. Deep learning applications is very computationally extravagant for getting good features and find in many cases that little data cannot give good learning features. We have used and reworked the architecture of Siamese Neural Network for One-shot learning to recognize the Bangladeshi bank notes with a tiny dataset. In this article we analyzed 20 images only for 5 different Bangladeshi bank notes what people used regularly. This research will help general people to get better experience with vending machines which can recognize notes with one data example only. We used 5 notes (5,10,20,50,100 TAKA) and get excellent result with 97.38% accuracy using help of convolutional architecture.