Automated Blood Group Detection Using Computer Vision and Transfer Learning on Mobile-Optimized Deep Neural Networks

This research presents an AI-assisted automated blood group detection framework using computer vision and MobileNetV2-based transfer learning for point-of-care biomedical diagnostics. The proposed system combines COCO annotation-driven region extraction, OpenCV preprocessing, and lightweight deep learning classification to identify agglutination patterns across Anti-A, Anti-B, and Anti-D reagent zones from full blood-card images. The dataset contains 7,836 annotated blood-card images:• Training Images: 6,456• Validation Images: 997• Test Images: 383 The model achieves:• ≈94–95% zone-level accuracy• ≈91–93% end-to-end ABO accuracy• Macro F1-score: 0.947 The system is optimized for lightweight mobile deployment using TensorFlow Lite and is designed as a research-stage assistive AI diagnostic framework for low-resource healthcare environments. This work does not claim clinical certification or replacement of laboratory blood typing procedures.