Pollution type sensing

A pollution sensing system comprising:

• a beam source positioned to direct a beam into an air sample from a monitored area;
• a photodetector positioned to sense light produced by interactions of the beam with particulate matter in the air sample;
• and
• a processing system connected to the photodetector, the processing system implementing:
• a data interface configured to produce raw data by sampling an output signal from the photodetector, wherein the raw data includes samples of measured light intensity from particle-beam interactions;
• a pulse evaluation module configured to:
• identify, in the raw data, pulses corresponding to a plurality of particles of the particulate matter, wherein each pulse of the pulses includes multiple samples for at least one particle and particle-beam interaction for the at least one particle;
• and,
• for each pulse, determine a characteristic of a particle corresponding to a pulse, wherein the characteristic corresponds to a size and/or albedo of the at least one particle based on width and/or height of the pulse;
• and
• a classification module configured to identify a type of a pollutant in the air sample through a classification process that analyzes data representing a distribution for characteristics of the plurality of particles,
• wherein the classification process includes:
• constructing a data vector from the raw data and distribution for the characteristics, wherein the data vector includes local particulate data, local non-particulate data, and contextual data, and the local particulate data includes distributions the size and the albedo based on pulse widths and pulse heights of a plurality of pulses;
• multiplying the data vector and a classification matrix, wherein the classification matrix is generated through a machine learning process trained using a plurality of air samples respectively containing a plurality of different types of pollutants, and the classification matrix encodes data regarding local particulate data, local gas data, and contextual Internet data for the plurality of different types of pollutants;
• and
• classifying the pollutant based on a result of the multiplying of the data vector and the classification matrix.

A process for identifying a pollutant in a monitored area, the process comprising:

• passing air from the monitored area through a sample cavity;
• directing a light beam into the sample cavity;
• sampling, as a plurality of samples, an output signal from a photodetector that is positioned to measure light produced from interactions of the light beam with particulate matter in the air passing through the sample cavity, wherein the sampling is at a frequency that produces multiple samples for each particle in a plurality of particles of the particulate matter;
• identifying, in the plurality of samples, a plurality of pulses, the plurality of pulses respectively corresponding to the plurality of particles;
• for each of the plurality of pulses, using a set of samples corresponding to a particle and a pulse to determine a characteristic of the particle, wherein the characteristic corresponds to a size and/or albedo of the particle based on width and/or height of the pulse;
• including, in a data set, a representation of a distribution of the characteristics of the plurality of particles;
• and
• identifying a type of the pollutant in the air through a classification process that analyzes the data set,
• wherein the classification process includes:
• constructing a data vector from the data set and the distribution for the characteristics, wherein the data vector includes local particulate data, local non-particulate data, and contextual data, and the local particulate data includes distributions of the size and the albedo based on pulse widths and pulse heights of a plurality of pulses;
• multiplying the data vector and a classification matrix, wherein the classification matrix is generated through a machine learning process trained using a plurality of air samples respectively containing a plurality of different types of pollutants, and the classification matrix encodes data regarding local particulate data, local gas data, and contextual Internet data for the plurality of different types of pollutants;
• and
• classifying the pollutant based on a result of the multiplying of the data vector and the classification matrix.