Slice-Inference-Assisted Lightweight Small Object Detection Model for Holographic Digital Immunoassay Quantification

Abstract

Sensitive and cost-effective detection methods utilizing portable equipment are crucial for applications in food safety inspection, environmental monitoring, and clinical diagnosis. In this study, we propose a sliced inference-assisted lightweight small object detection model (SIALSO) holographic biosensor for digital immunoassay-based quantification of chloramphenicol in food samples. This innovative biosensor combines a lens-free holographic imaging system with a lightweight deep learning model, capitalizing on the extensive field of view (FOV) of holography to facilitate precise signal detection of microsphere probes. The SIALSO model integrates a sliced inference-assisted algorithm to improve small object detection accuracy while minimizing computational complexity. Experimental results reveal that the SIALSO biosensor achieves a linear detection range from 50 pg/mL to 100 ng/mL (R² = 0.986), outperforming ELISA in both sensitivity and detection range. Furthermore, the model reduces computational parameters by 29% compared to YOLOv5s while maintaining high precision (98.2%) and recall (95.7%). This research establishes a robust theoretical and technological foundation for the development of portable detection devices in food safety and environmental monitoring.