A method for polarization detection and property identification of fine particulate matter based on LSTM

The accurate classification and property identification of fine aerosol particles are of great significance for atmospheric pollution source apportionment, environmental health assessment, and climate model optimization. Polarization measurement technology based on light scattering has become an important means for online dynamic characterization due to its high sensitivity to the microscopic properties of aerosols (particle size, complex refractive index, morphology). This paper combines machine learning to collect aerosol scattering data using a multi-angle polarization vector measurement device, constructs a spatial-polarization correlation matrix (angle × polarization state) as the input for a Long Short-Term Memory (LSTM) network, and uses the LSTM model to learn the temporal characteristics of scattering signals and the spatial-polarization correlation properties. The results show that the LSTM model demonstrates high application potential in the material classification of fine aerosol particles, providing new research ideas and methodological references based on temporal modeling for this field.