Precise and fast microdroplet size distribution measurement using deep learning

Abstract

Microfluidic technology is a versatile approach to improve the production of various fine chemicals and materials, where precise and fast size measurement of microdroplets in microscopic images is of pivotal significance for microfluidic device design. With recent developments in convolutional neural networks, we herein proposed a state-of-the-art deep learning-based method to cope with microdroplet size measurement and largely ease manual workload. The proposed method instance-wisely segments microdroplets with deep learning, and then fits their boundaries to obtain precise size distribution curves. Even overlapped droplets and small-sized satellite droplets can be detected and measured, which is not achievable in previous computer methods. Incredibly, diameter measurement error is as small as 0.75 μm, and the measurement efficiency is increased ~1000 times compared to manual measuring. This work not only sheds light on intellectual size measurement of microdroplets, but also points out a new way to promote microfluidic technology through deep learning methods.