This work introduces a method of continuous particle separation through standing surface acoustic wave (SSAW)-induced acoustophoresis in a microfluidic channel. Using this SSAW-based method, particles in a continuous laminar flow can be separated based on their volume, density and compressibility. In this work, a mixture of particles of equal density but dissimilar volumes was injected into a microchannel through two side inlets, sandwiching a deionized water sheath flow injected through a central inlet. A one-dimensional SSAW generated by two parallel interdigital transducers (IDTs) was established across the channel, with the channel spanning a single SSAW pressure node located at the channel center. Application of the SSAW induced larger axial acoustic forces on the particles of larger volume, repositioning them closer to the wave pressure node at the center of the channel. Thus particles were laterally moved to different regions of the channel cross-section based on particle volume. The particle separation method presented here is simple and versatile, capable of separating virtually all kinds of particles (regardless of charge/polarization or optical properties) with high separation efficiency and low power consumption.

No abstract

Removing undesirable reflections from a single image captured through a glass window is of practical importance to visual computing systems. Although state-of-theart methods can obtain decent results in certain situations, performance declines significantly when tackling more general real-world cases. These failures stem from the intrinsic difficulty of single image reflection removal -the fundamental ill-posedness of the problem, and the insufficiency of densely-labeled training data needed for resolving this ambiguity within learning-based neural network pipelines. In this paper, we address these issues by exploiting targeted network enhancements and the novel use of misaligned data. For the former, we augment a baseline network architecture by embedding context encoding modules that are capable of leveraging high-level contextual clues to reduce indeterminacy within areas containing strong reflections. For the latter, we introduce an alignment-invariant loss function that facilitates exploiting misaligned real-world training data that is much easier to collect. Experimental results collectively show that our method outperforms the state-ofthe-art with aligned data, and that significant improvements are possible when using additional misaligned data. arXiv:1904.00637v1 [cs.CV]