Dataset in support of the publication 'Classifying paint colour using acoustic data from laser ablation'

This dataset contains:

Fig. 1. Femtosecond laser pulses were focussed onto the surface of a dried painted region and the acoustic signal produced as a result of ablation was recorded using a USB microphone. The acoustic signals were then transformed into spectral images that were then used as the input to a neural network that classified the colour or the tone. The white bar indicates 20 µm in the inset ablated colour samples.

Fig. 2. Diagram showing examples of acoustic spectra for the different categories, for a) colour and b) tone (where 0 corresponds to white and 100 to black).

Fig. 3. The concept of using neural networks to identify (a) colour and (b) tone, from the acoustic spectra.

Fig. 4 (a) Confusion matrix showing the true class vs predicted class for the colour neural network being applied to acoustic spectra from the coloured samples. The blue shows the correct predictions and the orange the incorrect predictions, with the number of images tested on labeled in the corresponding squares. (b) Grad-CAM overlay of spectra from different samples, with the peak frequency (kHz) labeled in the inset in white.

Fig. 5 Testing of the tone neural network on data not used in training. Red circles represent test data 1, covering the full range from 0 to 100, while blue circles represent test data 2, comprising the entire categories of 30 and 70 that were excluded from training.