Machine learning techniques have been developed to identify inclusions on the surface of freely suspended smectic liquid crystal films imaged by reflected light microscopy. The experimental images are preprocessed using Canny edge detection and then passed to a radial kernel support vector machine (SVM) trained to recognize circular islands and droplets. The SVM is able to identify these objects of interest with an accuracy that far exceeds that of conventional tracking software, especially when the background image is non-uniform or when the target features are in close proximity to one another. This method could be applied to tracking objects in a variety of visually inhomogeneous fluid environments, in order to study growth dynamics, the development of spatial order, and hydrodynamic behavior.
Read full abstract