Machine learning classification of speckle patterns for roughness measurements

Classification based on Machine Learning (ML) algorithms has received particular attention in many fields. By associating speckle patterns of depolarized light produced by laser illumination of metallic rough surfaces and ML classification, we performed a study on roughness identification and its measurement. We obtained speckle patterns from metallic surfaces with different roughness, then we used the moments of a statistical distribution to apply neural network classification and the k-means clustering approach, i.e., supervised and unsupervised learning algorithms, respectively. We succeeded in finding a clear classification of speckle patterns concerning the roughness of the surfaces for two distinct roughness ranges This result has potential applications for surface monitoring. In addition, by using statistical parameters, we present a preliminary study on the potential use of such parameters to perform an indirect measurement of roughness. •Supervised and unsupervised Machine Learning to study speckle patterns classification from roughness surfaces.•Characterization of speckle patterns by using the moments of the intensity speckle distribution.•Surface roughness classification with a simple experiment.•Proof of principle for roughness characterization using k-means clustering approach and a calibration curve.