DEEP LEARNING-BASED ROBUST INDIAN COIN IDENTIFICATION IN RETAIL AUTOMATION

In modern financial ecosystems, the manual sorting and counting of coins remain a labor-intensive and error-prone endeavor. Traditional automation methods relying on weight sensors or basic image processing like edge detection and template matching frequently fail under real-world constraints such as variable lighting, coin wear, and complex backgrounds. The research addresses these inefficiencies by introducing an intelligent coin recognition system specifically designed for Indian currency denominations. This work evaluates a spectrum of computational approaches, starting with baseline machine learning models, including Multinomial Naïve Bayes (MNB), Decision Tree (DTC), and Random Forest (RFC). While these models yielded moderate success (70–85% accuracy), they struggled with the subtle feature variations of circulated coins. To achieve industrial-grade reliability, a Deep Convolutional Neural Network (CNN) architecture was developed. By extracting hierarchical spatial features and incorporating adaptive feature enhancement, the proposed deep learning model achieved a definitive 100% classification accuracy. The system is deployed via a dual-access Tkinterbased Graphical User Interface (GUI), featuring an ADMIN module for model orchestration (training/evaluation) and a USER module for real-time prediction and visualization. Performance was rigorously validated using precision, recall, and F1-score metrics. This framework provides a scalable, non-intrusive solution for banking and retail automation, effectively transforming currency management from a manual task into a seamless, automated process.