Thin cuvette speckle sample movies

Published: 2 November 2020| Version 1 | DOI: 10.17632/wnmn2pkg2s.1
Daniel Jakubczyk


Sample movies in RAW6 format of the dynamic speckle patterns from spherical nanoparticles suspensions in a thin (0.11 mm) cuvette. The complete set (as of Nov 2, 2020) consists of 639 files, which gives ~ 320000 scattering images. The images were used to train, validate and test neural networks (classifiers) in order to automatically recognise suspensions. The movies are hardly recognisable to the human experimenter, while the convolution network performed very well. Observation at 26.9 deg (from the forward). Illumination with polarised 532 nm light, 1.7 mm collimated beam. Camera Pike F-032C, AVT. In order to make use of the 14-bit raw images, a dedicated codec is required. Filters converting RAW6 to 24-bit RGB are provided by the camera manufacturer. Experimental details in the paper to be published. The files are labelled with substance name, nanoparticles size and suspension dilution. Every movie is ~500 frames long and was recorded at 80 fps. Since scattered light intensity differed by several orders of magnitude between the samples, the exposition (and if necessary – the camera gain) had to be adjusted. Only several discrete values of these were used. In case of doubt, whether the dynamic range was fully utilised, the measurement was repeated for a neighbouring exposition/gain value. Changing the camera parameters between the experimental runs – in particular, the exposition (image integration time) and gain (also amplifying the electronic noise) – may introduce non-meritorical information into the pictures and mislead/hinder the learning. Some difference of brightness between movies (series of images) seems unavoidable and was coped with at the data preparation stage. _v... indicates the number of the location in the cuvette



Polska Akademia Nauk Instytut Fizyki


Physics, Optics, Machine Learning, Scattering