Image classifier for the TJ-II thomson scattering diagnostic: Evaluation with a feed forward neural network

There are two big stages to implement in a signal classification process: features extraction and signal classification. The present work shows up the development of an automated classifier based on the use of the Wavelet Transform to extract signal characteristics, and Neural Networks (Feed Forward type) to obtain decision rules. The classifier has been applied to the nuclear fusion environment (TJ-II stellarator), specifically to the Thomson Scattering diagnostic, which is devoted to measure density and temperature radial profiles. The aim of this work is to achieve an automated profile reconstruction from raw data without human intervention. Raw data processing depends on the image pattern obtained in the measurement and, therefore, an image classifier is required. The method reduces the 221.760 original features to only 900, being the success mean rate over 90%. This classifier has been programmed in MATLAB.