Context and Challenges
Our client operates across multiple countries, handling various currencies with a proprietary solution for banknote recognition. Their system identified denominations, assessed banknote wear and tear, decoded serial numbers, and detected counterfeits.
However, their traditional computer vision approach faced key limitations:
- Time-consuming tuning: each new currency series or market required extensive manual calibration;
- Uncertain results: traditional methods struggled to adapt to new banknote designs, making the system less reliable in evolving markets;
- Deep learning challenges: while AI promised more adaptive solutions, the industry’s zero-error tolerance and the probabilistic nature of machine learning posed additional risks.
Solution
We redesigned the client’s banknote analysis pipeline, integrating a combination of image processing, machine learning, and deep learning methods. Our goal was to reduce reliance on hard-to-set thresholds and create a more flexible, robust system.
Key innovations included:
- Anomaly detection for counterfeit identification: To overcome the scarcity of counterfeit examples, we reframed the task as an anomaly detection problem. This approach not only improved detection but also ensured adaptability to future counterfeit variations.
- Hybrid AI models: We blended traditional computer vision techniques with AI models, striking a balance between reliability and adaptability.
- Hardware optimization: The final solution was carefully optimized to run seamlessly on the client’s existing hardware platform, eliminating the need for design changes, even for the deep learning models.
Results
Our solution enabled the client to accelerate model deployment for new currencies, reducing time-to-market. It boosted counterfeit detection accuracy while maintaining robust wear assessment. Most importantly, it ensured flawless denomination classification: after processing tens of millions of banknotes in the field, the system has yet to record a single misclassification. By integrating AI into their workflow, the client now benefits from a future-proof system, ready to tackle both emerging markets and evolving threats without compromising performance.
