Classification Rejection by Prediction

We address the problem of autonomous decision making in classification of radioastronomy spectrograms from spacecraft. It is known that the assessment of the decision process can be divided into acceptation of the classification, instant rejection of the current signal classification, or rejection of the entire classifier model. We propose to combine prediction and classification with a double architecture of Time Delay Neural Network (TDNN) to optimize a decision minimizing the false alarm risk. Results on real data from URAP experiment aboard Ulysses spacecraft show that this scheme is tractable and effective.