READ ME File For 'Dataset for Determination of size of urban particulates from occluded scattering patterns using deep learning and data augmentation'

Dataset DOI: 10.5258/SOTON/D1668

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, Matthew Praeger, Matthew Loxham, Robert W. Eason and Ben Mills
TITLE:Determination of size of urban particulates from occluded scattering patterns using deep learning and data augmentation
JOURNAL: IOP Environmental Research Communications
PAPER DOI IF KNOWN: 10.1088/2515-7620/abed94


This dataset contains:

Fig1.jpeg
Fig2.jpeg
Fig3.jpeg
Fig4.jpeg
Fig5.jpeg
Fig6.jpeg

Fig3a.txt Data for X-dimension.
Fig3b.txt Data for Y-dimension.
Fig4a.txt Data for Y-dimension comparison.
Fig5.txt Data for Y-dimension comparison with augmentation.
Fig6a.txt Data for the RMSE in Y-dimension.

The figures are as follows:
Fig. 1 (a) Illustration of experimental setup for using a camera to collect scattering patterns from urban particulates deposited onto a microscope slide, when the particulates were illuminated with red, green and blue laser light. The particulates were also imaged for reference and sample alignment. (b) Scattering patterns and, corresponding SEM images and optical images for three different particulates (upper, middle and lower rows, respectively). The SEM images include labels of the maximum X-dimension and maximum Y-dimension of the particulates.

Fig. 2 (a) Illustration of the concept of the training of a neural network using a scattering pattern paired with a dimension (e.g. Y = 15 µm), and then testing the network on a scattering pattern. Here, the test scattering pattern includes black circles (zero intensity pixel occlusions) to simulate particulates occluding the sensor.

Fig. 3  Capability of the trained neural networks to determine the size of the urban particulates for (a) X-dimension and (b) Y-dimension. 

Fig. 4  (a) Difference between the prediction of the Y-dimension for particulate scattering patterns without occlusions and scattering patterns containing occlusions, for a total of 1 (red), 10 (green), 100 (blue) and 1000 (yellow) randomly located single occlusion pixels. (b) Scattering patterns and maximum activation of the neural network for 8 µm and 25 µm with zero pixel occlusions. 

Fig. 5  Difference between the prediction, using the neural network trained on augmented data, of the Y-dimension for particulate scattering patterns without occlusions and scattering patterns containing occlusions with single pixel width, for a total of 1 (red), 10 (green), 100 (blue) and 1000 (yellow) randomly located occlusion pixels.

Fig. 6  (a) Absolute difference between the RMSE of all the tested scattering patterns (5-25 µm) predictions by neural network trained on original unedited data (solid circles) and a neural network trained on augmented data (hollow circles), compared with the RMSE predictions without occlusions. (b) Scattering patterns showing example occluded scattering patterns covered randomly in 100 zero intensity pixel circle occlusions of 1, 5 and 10-pixel diameters.



Date of data collection: 20/09/2019

Information about geographic location of data collection: Building 46, University of Southampton
 
Licence: CC-BY

Related projects:
EPSRC grant EP/N03368X/1
EPSRC grant EP/T026197/1
BBSRC grant BB/P011365/1


Date that the file was created: 03, 2021