Broad Learning System with Preprocessing to Recover the Scattering Obstacles with Far-Field Data

Based on Broad Learning System with preprocessing, the impenetrable ob-stacles were reconstructed. Firstly, the far-field data were preprocessed by Random Forest, and the shapes of the obstacles were classified by dividing the far-field data into different categories. Secondly, the broad learning system was employed for re-constructing the unknown scatterer. The far-field data of the scatterer were regarded as the input nodes of mapped features in the network, and all the mapped features were connected with the enhancement nodes of random weights to the output layer. Subsequently, the coefficient of the output can be obtained by the pseudoinverse. This method for the recovery of the scattering obstacles is named RF-BLS. Finally, numer-ical experiments revealed that the proposed method is effective, and that the training speed was significantly improved, compared with the deep learning method.