NOISE REMOVAL IN IONOGRAMS BY NEURAL-NETWORK

A neural network approach to the extraction of 1-hop F-layer traces fr om oblique-incidence ionograms is shown to offer performance at beast comparable with conventional filtering techniques. Preprocessing in th e form of background noise and vertical (horizontal) line removal is u tilised prior to training 110:7:100 MLP using backpropagation with mom entum. It is further demonstrated that such successful trace extractio n can be achieved with just 50 ionogram training exemplars.