Application of artificial neural network (ANN) technique to rough surface inverse scattering problems

Summary form only given. The authors introduced an ANN multilayered perceptron technique and its application to rough surface inverse scattering problems. For the rough surface scattering problems considered, the values of the spectral correlations of the scattered intensities are the input and the desired roughness parameters, the RMS height, and the correlation distances are the output. The authors solved the forward problem, applied the input, and used arbitrarily chosen initial weights and bias to obtain the output. They then found the error between the target and the output. They used the backpropagation technique to change the weight and bias to reduce the error. The above learning procedure was repeated many times to train the ANN. Once the ANN was trained, the measured inputs were applied to the ANN and the desired outputs were obtained quickly. Next the inverse problem of finding the surface parameters, sigma and l, from the measurement was considered.<>