Improvement of identification accuracy of multisensor conversion characteristic using SVM

A method of individual conversion characteristic identification of multisensor using reduced number of its calibration/testing results is described in this paper. The proposed method is based on the neural-based reconstruction (approximation or prediction) of surface points of multisensor conversion characteristic. Each neural network module reconstructs separate point of the surface. Our results show that the use of a Support Vector Machine (SVM) model allows improving the reconstruction accuracy of multisensor conversion characteristic. The reconstruction results obtained by SVM are compared with the results obtained by a multi-layer perceptron (MLP).