Using neural networks to analyse surface irregularities measured with holographic radar

Holographic radar is very sensitive to small irregularities in surface height [1][2]. Although this sensitivity was previously thought a disadvantage of holographic radar, recent measurements on dinosaur footprints [3] have shown that it can provide valuable information. A problem has been that the RASCAN 4 holographic radar system used in this work has provided separate signals at 5 different frequencies between say 3.6 and 4.0 GHz and at both parallel and perpendicular polarisations, each of which gives a distinct signal as a function of surface height and other variable. These signals are complicated to calculate but can be measured using a sloping surface of known height and other properties. Here neural networks are trained on a gently sloping surface of smooth sand to recognise the RASCAN signals as a function of surface height. In a testing mode, the neural networks should be able to use all the recorded signals to distinguish small differences in surface height as a function of position.