READ ME File For 'Airborne Particulate Matter Sensing via Laser Filament–Interaction and Deep Learning'

Dataset DOI: hhttps://doi.org/10.5258/SOTON/D3818

ReadMe Author: James A Grant-Jacob  University of Southampton https://orcid.org/0000-0002-4270-4247

This dataset supports the publication:
AUTHORS: James A. Grant-Jacob and  Ben Mills
TITLE: Airborne Particulate Matter Sensing via Laser Filament–Interaction and Deep Learning
JOURNAL: ACS ES&T Air
PAPER DOI IF KNOWN:  


This dataset contains:

Figure_Images.zip
(Figure_1.jpg 
Figure_2.jpg
Figure_3.jpg 
Figure_4.jpg
Figure_5.jpg 
Figure_6.jpg)

Figure_Data.zip
Figure2_shape_size_summary.txt)

The figures are as follows:

Figure 1. (a) Experimental setup for filament-based airborne particulate detection. Schematic showing a 1030 nm 190 fs laser pulses focused using a 100 mm focal length lens into ambient air to generate a laser filament. Airborne particulates (Chalk, Pollen and Salt) are delivered into the beam path via a funnel placed 500 µm above the filament. Optical emission events were imaged orthogonally using a Basler RGB camera (daA1920-160uc, 1920 × 1200 pixels) coupled with a 50× Olympus long working distance objective (SLMPLN, 0.35 NA). (b) Plasma channel images at increasing power, showing extension, splitting and intensity growth with increasing pulse energy. (c) Spectra for different laser powers taken for 5 second integration times showing broadening beyond 550 nm and onset of filamentation at ~85% power. (d) Examples of microscope images of Blank (i.e., nothing), Chalk, Pollen and Salt, along with example of a corresponding filament image with optical emission events being fed into the neural network for prediction. Note the brightness of the filament images have been increased for ease of viewing.

Figure 2. (Left) Scatter plot of diameter versus circularity for Chalk, Pollen and Salt particles. (Right) Boxplots showing size variability per class with mean and standard deviation annotations. Pollen exhibits higher circularity and lower eccentricity compared to Chalk and Salt, consistent with its more spherical morphology.

Figure 3. Architecture of the CNN trained and implemented in MATLAB.

Figure 4. Confusion matrix of the results of classification for the neural network testing on (a) Images of optical emission events at 100% laser power (b) Images of optical emission events at 85% laser power.

Figure 5. (a) Images of optical emission events for Blank (filament only), Chalk dust, Pollen grains Salt crystals captured in the filament region and their corresponding (the intensity of the brightness of the images has been increased for ease of viewing). (b) Grad-CAM heatmaps highlighting the most influential regions in the CNN model’s prediction.

Figure 6: (a) 2D scatter plot of chromaticity values (Red versus Green) extracted from optical emission events. Each point represents the chromaticity coordinates computed from the top 30 brightest pixels within Grad-CAM–highlighted regions of the original images. Colours indicate particle classes, illustrating the separation and overlap in chromaticity space between Blank, Chalk, Pollen, and Salt. (b) Representative Grad-CAM patches for each class, illustrating that Chalk often has bright pink features, Pollen as distinct green spots, Salt as diffuse orange–green patches, and Blank frames as dark with minimal signal. (c) Corresponding emission spectra for particles from each class, acquired using 1 s integration times and dark background subtraction.  


The graph data are contained in the tab separate data files as follows, with their relevant headings:

Figure2_shape_size_summary.txt

License: CC-BY

Related projects:
EPSRC grant EP/T026197/1
EPSRC grant EP/W028786/1
EPSRC grant EP/Z002567/1
Date that the file was created: 10, 2026