Nepali Cash Detection and Recognition
Currency detection and classification for Nepali banknotes using InceptionV3 transfer learning, achieving 94% accuracy across 7+ denominations.
Problem
Automating Nepali banknote identification for assistive technology and cash-handling requires robust classification across denominations with limited training data.
Approach
Transfer learning with InceptionV3 pretrained on ImageNet: freeze the convolutional backbone, train a custom classification head with augmentation.
Architecture
InceptionV3 (frozen backbone) → GlobalAveragePooling → Dense(256, ReLU) → Dropout(0.5) → Dense(7+, Softmax).
Results
94% accuracy on test and validation. Balanced performance across all denominations. Expanded in 2023 update.
Lessons learned
Transfer learning from ImageNet generalizes remarkably well to domain-specific tasks with limited data — pretrained feature reuse outweighs distribution mismatch.
Next
Fed-BLEND — Federated Conformal Prediction for VLMs
Novel federated conformal prediction method that mitigates hallucinations in federated fine-tuned vision-language models via abstention.
GCTAF — Time-Series Forecasting & Flare Risk Classification
Attention-based forecasting of magnetic field trajectories combined with supervised contrastive learning for solar flare classification.