A Deep Learning Approach for the Automatic Video Classification of Silicone Oil Droplet Deformation Induced by Electrohydrodynamic Effect

Various body fluids exhibit altered rheological properties in the presence of diverse pathologies. Although biofluid viscosity could be a potential diagnostic marker, the large amount of samples required by traditional viscometers limits its use in clinical setting. Hence, there is a significant interest to retrieve rheological characteristics of such liquids at the microscale by means of compact, intelligent, and reliable platform. In this context, this paper introduces an intelligent system based on advanced deep learning (DL) techniques which is able to extract automatically spatial-temporal dynamics during the process of Deformation of a liquid droplet in response to the electro-hydrodynamic (EHD) pressure, captured by a suitable side-view video imaging system. In particular, the morphological changes of silicone oils with different viscosity are analysed. The developed DL-framework exploited the pretrained GoogleNet to extract spatial features, that are subsequently used as input to a bidirectional long short-term memory (BiLSTM) network to learn temporal dynamics across frames and perform the related binary video classification between two different silicone oil viscosity (10 Pa.s vs. 5 Pa.s). Experimental results encourage the use of the proposed framework also for future biofluids classification and rheological characterization.