Anomaly size estimation by neural networks based on electrical impedance tomography boundary measurements

A previously proposed approach based on RBF neural networks for detecting anomaly location is extended to estimate the anomaly size. First, a predefined number of threshold values are selected in the range of possible anomaly sizes. Next, RBF neural networks are used as classifiers to classify the anomaly size as being smaller or larger than each threshold value. The inputs of the classifiers are the data obtained from EIT boundary measurements. The anomaly size can be estimated by properly cascading the classifiers. The estimation precision is adjusted by the number of threshold values.