Scholarlink

<p>Digital microfluidics (DMF) enable programmable manipulation of droplets, but the inevitable generation of microscopic particles necessitates the isolation of particle-free droplets for downstream analysis. Manual inspection is labor-intensive and impractical for high-throughput scales. We propose an integrated DMF system featuring a multi-scale object detection model based on YOLOv8 to achieve automated particle identification and droplet manipulation. Our model incorporates higher-resolution detection layers and an Efficient Local Attention module to overcome challenges posed by extreme scale differences and dense spatial distributions. Experimental results demonstrate that the model achieves a mean Average Precision (mAP@0.5) of 96.9% and a low counting error (MAE = 3.6) across various particle densities. Furthermore, we successfully implemented a real-time, visually guided workflow for automated splitting of particle-free droplets with a success rate of 94.0%. This system provides a robust foundation for automated quality control in high-throughput DMF-based biochemical assays.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

An intelligent multi-scale detection system for particle detection and automated splitting of particle-free droplets in digital microfluidic systems

  • 期刊论文

  • 发表日期：

    2026年4月13日

查看全文
  • Kunlun Guo,
  • Juyue Dong,
  • Junyu Zhang,
  • Zerui Song,
  • Huifeng Wang

摘要

Digital microfluidics (DMF) enable programmable manipulation of droplets, but the inevitable generation of microscopic particles necessitates the isolation of particle-free droplets for downstream analysis. Manual inspection is labor-intensive and impractical for high-throughput scales. We propose an integrated DMF system featuring a multi-scale object detection model based on YOLOv8 to achieve automated particle identification and droplet manipulation. Our model incorporates higher-resolution detection layers and an Efficient Local Attention module to overcome challenges posed by extreme scale differences and dense spatial distributions. Experimental results demonstrate that the model achieves a mean Average Precision (mAP@0.5) of 96.9% and a low counting error (MAE = 3.6) across various particle densities. Furthermore, we successfully implemented a real-time, visually guided workflow for automated splitting of particle-free droplets with a success rate of 94.0%. This system provides a robust foundation for automated quality control in high-throughput DMF-based biochemical assays.