AI-driven digital inline holography for real-time in situ particle analysis

Real-time in situ analysis of particles (e.g., size, concentration, and shape) is crucial for environmental monitoring, medical examination, advanced manufacturing, and many other areas. The digital inline holography (DIH) has emerged as a low-cost imaging-based solution for high-fidelity in situ measurements of particle shape and types (in addition to size and concentration) which are not available from conventional particle sensors based on light scattering or aerodynamic characteristics. However, the existing DIH method is computationally expensive, conducted offline and lacks robustness to deal with changing image quality (due to fluctuation in background intensity and noise, etc.) present in in situ applications. To address these challenges, we introduce a machine learning framework for simultaneous detection, classification, and other analyses (e.g., viability) of particles of various forms in real-time. The framework involves a combination of different convolutional neuron networks and generative adversarial networks for synthesis, autonomous labeling, and processing of DIH data. Integrated with various hardware setups, the performance of this AI-driven DIH approach has been demonstrated in monitoring of indoor air quality, harmful algal blooms and cell sorting.