Deep Learning-Based Droplet Menisci Recognition for Digital Microfluidic Devices

This paper presents the application of machine learning techniques, specifically deep learning, to enhance the performance and reliability of electrowetting-on-dielectric (EWOD) digital microfluidic (DMF) devices. During droplet manipulation within an EWOD DMF device, the menisci of the droplets carry valuable information, including fluid properties, droplet size, position, and dynamics. Recognizing and analyzing these droplet menisci can provide insights into reaction kinetics and device defects, enabling reliable feedback control for improved device operation. To achieve reliable control of EWOD DMF devices, this study focuses on the recognition of droplet menisci using a deep learning approach. The U-Net model, a highly effective deep learning architecture for image segmentation tasks, is integrated into the present study. A convolutional neural network with the U-Net architecture is implemented, and image segmentation is performed to accurately identify droplet menisci. The network is trained using a diverse dataset containing images of droplet under various conditions. Model parameters are optimized to ensure high accuracy and robust performance. Experimental tests are conducted on images of droplets in different conditions, and the trained network successfully recognizes menisci with a model accuracy of 98 %.