Recognition of earthquakes and explosions using a data compression neural network

The ability to reconstruct unlearned images using a neural network trained with a learned (known) image represents the generalization property of the network. The degradation of the reconstructed image compared to the original is expected to be least for the reconstructed, learned image, somewhat greater for a similar but unlearned image, and significantly greater for a dissimilar unlearned image. The method developed and tested is based on the generalization properties of the trained image compression neural network and a measure of the degradation of the reconstructed image over the population of similar events and dissimilar events (i.e. explosions vs earthquakes)