UCLA

Air quality is an increasing concern for human health. Among all pollutants, bio-aerosols, which account for 5−34% of indoor particulate matter, are micro-scale airborne living organisms that originate from plants or animals. Inhaled by a human, they can cause irritation, allergies, and various diseases. Conventional detection of bio-aerosols still relies on a pipeline developed for more than 50 years: an aerosol sample is taken at the inspection site, then transmitted to a laboratory and inspected manually under a microscope or through culture experiments. There is still an urgent and unmet need for accurate, label-free and automated bio-aerosol sensing to cover a wide range of species, ideally within a field portable, compact, and cost-effective platform. To this end, we have developed a portable device for air quality monitoring, termed c-Air. It samples aerosols using an impactor, takes holograms of the collected particles using lens-free microscopy and classifies those unlabeled samples using deep learning tools. This device opens new opportunities for environmental sensing and personalized air quality monitoring. This dissertation aims to mature the c-Air device and extend its utility in different real-world applications. A deep learning-assisted workflow for label-free bio-aerosol detection was streamlined, including neural network-based holographic reconstruction and particle classification. The reconstruction network achieves auto focusing and phase recovery simultaneously, greatly reducing the computational complexity. Enabled by this high-efficiency phase recovery method, high throughput monitoring of phase-only, transparent particles becomes feasible. We upgraded the hardware design to monitor the dynamic change of volatile liquid samples such as electronic cigarette (e-cig) generated aerosols using time-lapsed images. Both in-lab and field experiments demonstrated a direct and high-throughput volatility quantification method of e-cig-generated aerosols. And this method can be broadly applied to rapidly characterize various volatile particulate matter. We also developed the next generation of the c-Air device, replacing impactors for particle collection with a virtual impactor to avoid overloading of the sampling cartridge and elongating its lifetime. Combining the virtual impactor with lens-free microscopy, the new device provided volumetric, time-lapsed measurements on flying aerosols. The reconstructed holograms can be further used for particle classification with a neural network. This AI-based bio-aerosol detection and classification device provides a unique solution to indoor air quality monitoring and label-free bio-aerosol sensing.