Domain check for input to NN emulating an inverse model

Multi-layer perceptrons are used in many fields of science and technology to emulate forward and inverse models. The paper proposes a necessary condition check for the validity of results obtained by inverse NNs. The solution consists of a combination of inverse and forward NNs that, in addition to the inversion, computes a quality measure for the inversion. Examples of such combined NNs from applications in the remote sensing area are discussed.