AI-Enhanced Virus Detection in Biopharmaceutical Production Processes

Ensuring viral safety is a fundamental requirement in the production of biopharmaceutical products. Transmission Electron Microscopy (TEM) has long been recognized as a critical tool for detecting viral particles in unprocessed bulk (UPB) samples, yet manual counting remains labor-intensive, time-consuming, and prone to errors. To address these limitations, we propose an enhanced virus strain detection approach using the YOLOv11 deep learning framework, optimized with C3K2, SPPF, and C2PSA modules in the backbone, PANet in the neck, and Depthiwise Convolution (DWConv) in the head. To further improve feature fusion and detection of single-class virus particles, we integrated BiFPN and C3K2_IDWC modules. The resulting model (YOLOv11n + BiFPN + IDWC) achieves an mAP@0.5 of 0.995 with 33.6% fewer parameters compared to YOLOv11n, while increasing accuracy by 1.3%. Compared to YOLOv8n and YOLOv10n, our approach shows superior performance in both detection accuracy and computational efficiency. These results demonstrate that the model offers a robust and scalable solution for real-time virus detection and downstream process monitoring in the pharmaceutical industry.