US20260086013
2026-03-26
Physics
G01N15/0227
The patent application describes a novel device for detecting particulate matter, particularly bio-aerosols, using holography and deep learning. The system employs a virtual impactor to concentrate larger particles by selectively slowing them down and guiding them through an imaging window. As the particles flow, they are illuminated by a pulsed laser diode, creating holographic images that are captured by a CMOS image sensor in a lens-free mobile device. The device is designed to be mobile, cost-effective, and capable of extended operation without the need for cartridges or filters.
This invention is situated in the field of bio-aerosol and particulate matter detection and analysis. It specifically addresses the need for portable devices that can sample and inspect aerosols in the field. Traditional methods involving physical sampling and laboratory analysis are complex and require skilled personnel, making them unsuitable for continuous monitoring. The new device overcomes these limitations by integrating aerosol sampling with advanced imaging and classification technologies.
Bio-aerosols, which include pollen, fungi, and bacteria, are a significant component of indoor particulate matter. These particles can cause health issues such as allergies and respiratory diseases. Conventional detection methods rely on lab-based analysis, which is time-consuming and requires expertise. Existing portable devices often use antibody-based detection, which is limited by storage issues and specificity. Virtual impactors have been used for separating particles based on inertia, but they typically require additional filtration steps for classification.
The device detailed in the application combines a virtual impactor with computational imaging and deep learning to classify bio-aerosols without labels or chemical processing. It uses a 3D-printed virtual impactor to concentrate particles larger than approximately 6 μm, and an inline holography setup to capture their images. The device captures three holograms per particle in a single frame, which are processed by a deep neural network to classify the particles. Testing with various pollen types achieved a high classification accuracy of 92.91%.
One embodiment includes a device with an air sampler, a virtual impactor, and an imaging window, where a pulsed light source and image sensor capture holographic images. The images are processed by a computing device to create focused images, which are then classified by a neural network. Another method involves drawing air through the device, capturing holographic images, and using deep learning for classification. The system can be adapted for different particle types and provides a scalable solution for air quality monitoring.