A Hybrid-Based Deep Learning Framework for Pill Detection 

Optimizing automated pill detection and multi-attribute classification in healthcare imaging presents significant challenges due to variations in pill shape, size, color, and packaging. This paper proposes a hybrid deep learning framework that integrates YOLOv26, RT-DETR, and Faster R-CNN to enhance detection, classification, and metadata retrieval for pharmaceutical datasets. The framework combines the high-speed localization capabilities of YOLOv26, the transformer-based contextual reasoning of RT-DETR, and the region proposal strength of Faster R-CNN to accurately identify pills, classify multiple attributes simultaneously, and retrieve associated metadata. Extensive experiments are conducted on a multi-class pill dataset, demonstrating the effectiveness of the hybrid approach in handling diverse visual characteristics and complex class distributions. Additionally, the framework is complemented by a scalable pipeline for integrating attribute-level predictions with metadata queries, facilitating advanced pharmaceutical analysis and inventory management. This study underscores the potential of hybrid architectures to unify complementary detection and classification strategies for complex biomedical imaging tasks.

Keywords: Pill Detection, Multi-Attribute Classification, YOLOv26, RT-DETR, Faster R-CNN, Metadata Retrieval, Deep Learning, Hybrid Framework, Object Detection, Biomedical Imaging