Design and Implementation of a High-Throughput Digital Microfluidic System Based on Optimized YOLOv8 Object Detection.

To address the challenges of excessive control pins and inefficient high-throughput droplet manipulation in conventional digital microfluidic chips, this study developed a parallel-motion digital microfluidic system integrated with an image acquisition device. The system employs an enhanced YOLOv8 object detection model for droplet recognition. By enabling parallel droplet transportation and processing, it significantly improves operational efficiency and detection accuracy. For droplet recognition, the YOLOv8 model was optimized through the integration of GAM_Attention and EMA mechanisms, which strengthen feature extraction capabilities and detection performance. Experimental results demonstrated that the optimized model achieves remarkable accuracy and robustness in droplet detection tasks, with mAP50 increasing from 96.5% to 98.7% and mAP50-90 improving from 65.8% to 68.5%. The system exhibits enhanced detection precision and real-time responsiveness, maintaining an error rate below 0.53%. Furthermore, a host computer interface was implemented for multi-droplet path planning and feedback, establishing a closed-loop control system. This work provides an efficient and reliable solution for high-throughput operations in microfluidic chip applications.