The effectiveness of face masks for preventing airborne transmission has been debated heavily during the COVID-19 pandemic. This paper investigates the filtration efficiency for four different face mask materials, two professional and two homemade, for different airflow conditions using model experiments and artificially generated water droplets. The size range chosen represents particles with the largest volume that can be suspended in air. The particles are detected using double pulsed interferometric particle imaging, from which it is possible to estimate the positions, velocity, and size of individual particles. It is found that all the tested face masks are efficient in preventing particles from transmission through the mask material. In the presence of leakage, particles larger than approximately 100𝜇m are completely removed from the air stream. The filtration efficiency decreases with the decreasing particle size to approximately 80% for 15𝜇m particles. The size dependency in the leakage is mainly due to the momentum of the larger particles. The results show that even simple face mask materials with leakage prevent a large portion of the emitted particles in the 15–150 𝜇m range.

Three-dimensional particle positioning from inline holograms is performed using convolutional neural networks. The faster R-CNN architecture is implemented for multi-particle identification and lateral positioning, and a second network estimates the depth position. Supervised learning is used to train the network using simulated holograms.

In this paper we generate word meta-embeddings from already existing embeddings using cross-encoding. Previous approaches can only work with words that exist in each source embedding, while the architecture presented here drops this requirement. We demonstrate the method using two pre-trained embeddings, namely GloVE and FastText. Furthermore, we propose additional improvements to the training process of the meta-embedding. Results on six standard tests for word similarity show that the meta-embedding trained outperforms the original embeddings. Moreover, this performance can be further increased with the proposed improvements, resulting in a competitive performance with those reported earlier.

Dual-view polarization-resolved digital holography is used to record the polarization response of the scattered light from both spherical and non-spherical nanoparticles. A model based on the T-matrix method is developed to map the polarization to a size estimate of the volume equivalent radius.

In this paper, we demonstrate how polarization-resolved holography can be used to determine if a particle is spherical or not and to estimate the size information of nanoparticles. The T-matrix method is used to model the scattered light from both spheres and spheroids. A dual-view polarization-resolved imaging system is used in order to obtain polarization ratio angles (β₁,β₂). From the obtained β₁ and β₂, it is possible to estimate whether or not a particle is spherical or not. It is found that non-sphericity only has a minor effect up to around sizes of 120nm, and for that range, a spherical approximation is valid. For larger particles, the orientation influence the polarization response greatly. The size of a non-spherical particle can be estimated from the polarization ratio angles. The upper limit we can estimate unambiguously is around 200nm. Finally, the model is applied to experimental measurements of naturally occurring particles in purified water. From the measurements, it is possible to separate spherical from non-spherical particles and also give a rough estimate of the size.

In this thesis, measurement of particle fields using digital holography is the main subject. The questions investigated consider positioning, identification and sizing of nanometer and micrometer particles. The thesis explore these topics using both simulations and experimental measurements. Measuring particles is inherently a three-dimensional problem. Digital holography is, therefore, chosen as the measurement technique since it can record three-dimensional information in the interference pattern.

Two main digital holographic setups are considered in this work, one single-view and one dual-view, both with off-axis configuration for the reference wave. Methods for positioning along the optical axis is the central question for the single-view system. This work presents a new method for axial positioning based on the wavefront curvature of the scattered light. In the reconstructed volume, along the optical axis, the scattered wave changes from converging to diverging around its location. This assumption is verified using simulations. An estimation of the position where this change occurs, hence, is an estimation of the actual axial position. Two different methods for quantification of the wavefront curvature is presented. The first uses the finite difference method of the reconstructed phase. The second uses a Chebyshev model for the phase-response. The difference between the two methods is that the one based on the Chebyshev model is more robust and less sensitive to noise.

The dual-view system is an extension of the single-view setup where an identical system is placed perpendicular with the first system. The sample is illuminated from below, making the angle between the illumination and the two systems $90^\circ$. The concept of polarization-resolved registration is also incorporated in the detection. This detection is made possible by using two reference waves with linear and mutually orthogonal polarization at different off-axis tilts. One hologram can, therefore, be reconstructed into two complex amplitudes, one for each polarization component. The measurements using this system focus on how particle properties influence the polarization-response. The transition from single to dual-view polarization-resolved detection increases the complexity in the reconstruction. There is a need for accurate calibrations for this type of setup. The thesis contains details on calibrations of both the spatial mapping and polarization detection.

The first application of the dual-view polarization-resolved setup is for identification and size estimation of nanometer-sized particles. The T-matrix method is used to establish a model for the sizing of spherical and spheroid particles. It is possible to estimate the size unambiguously up to approximately 200 nm for smooth particles. The sizing is limited to this lower size range.

The second application investigates how the detected polarization varies for different kinds of microplastics. The measurements show that the microplastics have a complex polarization-response, indicating an irregular and non-spherical shape. In general, for both sizing and identification, the problem becomes more complex as the size increases and the particle shape is less smooth.

This thesis shows that it is possible to estimate particle information from the polarization-response. However, the method is constricted both by the size and complexity of the particles.

A dual-view polarization resolved digital-holographic system is presented. The necessary calibration for both polarization and spatial coordinates are outlined. As an example the system is is used to track spherical microparticles in a cuvette.

A novel dual-view polarization-resolved pulsed holographic system for particle measurements is presented. Both dual-view configuration and polarization-resolved registration are well suited for particle holography. Dual-view registration improves the accuracy in the detection of 3D position and velocities, and polarization-resolved registration provides polarization information about individual particles. The necessary calibrations are presented, and aberrations are compensated for by mapping the positions in the two views to positions in a global coordinate system. The system is demonstrated on a sample consisting of 7 μm spherical polystyrene particles dissolved in water in a cuvette. The system is tested with different polarizations of the illumination. It is found that the dual view improves the accuracy significantly in particle tracking. It is also found that by having polarization-resolved holograms, it is possible to separate naturally occurring sub-micrometer particles from the larger, 7 μm seeding particles.

When a particle is illuminated by a beam of light it will scatter and redistribute the light in all directions. How it scatters depends on the size, shape and refractive index of the particle. Additionally, it depends on the wavelength and polarization of the illuminating beam. The direction and distance to the observer relative the particle also needs to be considered.  A digital holographic imaging system is used to collect parts of the scattered light from micrometer-sized particles. By utilizing digital holography a three-dimensional reconstruction of the imaged scene is possible. Traditionally, particles are localized based on the intensity in the holographic reconstructions. In this licentiate thesis, the phase response of the scattered light is investigated and utilized. An alternative method for locating spherical particles is presented. The method locate particles based on a simple feature of a propagating wave, namely the fact that the wavefront curvature changes from converging to diverging at the axial location of the particle. The wavefront curvature is estimated using two different methods. The first method estimates the lateral phase-gradients using a finite-difference method. The second method uses a three-dimensional parametric model based on a Chebyshev polynomial expansion. The methods are demonstrated using both simulations and experimental measurements. The simulations are based on the Lorenz-Mie scattering theory for spherical particles and are combined with an imaging system model. Experiments are performed using an off-axis polarization sensitive digital holographic system with a coherent Nd:YAG laser. Measurements of stationary particles are made to validate and evaluate the proposed method. It is found that these methods estimate the true axial position and does not have the offset that is associated with intensity-based methods. Additionally, it is possible to exclude noise that shows up as false particles since noise does not have the same phase response as a real particle. The second method, that uses a parametric model, also improves the standard deviation in the positioning.