Efficient RCS Prediction of Composite Scene Based on Deep BP Neural Networks

Calculating composite electromagnetic scattering characteristics of the target and the environment is often performed by either experiments or simulations. However, the experiment is easily interfered by environmental factors, and electromagnetic simulation is faced with the problems of time consuming or limited accuracy. In this paper we propose a deep BP neural network for fast obtaining the composite scattering characteristics of target and environment. We develop a composite scattering database of ideal conductor cube target and dielectric slab to train the proposed deep BP neural network. The trained neural network quickly predicts the composite scattering of the cube target and the flat plate. The results confirm that the trained network can explain up to 98% of the training sample data. In addition to the training data, we tested the generalization ability of the neural network. We selected four groups of samples outside the database as examples to verify the data accuracy of neural network prediction. Then, the prediction results of the neural network are compared with the simulation results of commercial software FEKO. The results of the four sets of test data are as follows. The error of the first set of test data was less than 0.5 dB at all angles. The second test data showed that the Angle with an error of less than 0.5 dB accounted for 95.56% of all the predicted data. The third experimental data showed that the Angle with an error of less than 5 dB accounted for 97.77% of all the predicted data. The error of the fourth group of test data was less than 1 dB in all angles. The results suggest a high generalization capability of the neural network, where the predicted error is within the system design requirements. The work resented in this paper provides new ideas and directions for the calculation of complex scattering characteristics of more complex targets and actual scenes.