Application of neural nets to feature fusion

A multiple-feature-multiple-neural fusion solution to the problem of detecting and classifying transient signals, denoted as a hierarchical neural net, is described. The fusion of the multiple features leads to significant gain in both detection and classification performance while at the same time reducing false alarms. The multinet approach gives very good results even in cases where the first layer nets operating on single features fail. Results comparing the single feature nets and the hierarchical neural net approach processing real data are presented