Intelligent Optical Tweezers with deep neural network classifiers

Abstract

Optical tweezers use light to trap and manipulate mesoscopic scaled particles with high precision making them a useful tool in a plethora of natural sciences, with emphasis on biological applications. In principle, the Brownian-like dynamics reflect trapped particle properties making it a robust source of information. In this work, we exploit this information by plotting histogram based images of 250ms of position or displacement used as input to a Convolution Neural Network. Results of 2-fold stratified cross-validation show satisfying classifications between sizes or types of particles: Polystyrene and Polymethilmethacrylate thus highlighting the potential of CNN approaches in faster and non-invasive applications in intelligent opto and microfluidic devices using optical trapping tools.