An EM Scattering Data Interpolation Method Based on Decision and Regression Tree

The granularity of electromagnetic (EM) scattering data is a key aspect in studying targets’ EM scattering characteristics. The basic EM scattering data acquiring methods are doing simulation and measurement. For targets with unknown model parameters, e.g., geometric and material parameters, it’s hard to use direct simulation method to get their detailed EM data. Due to the high cost and limited measurement conditions, the acquired EM scattering data is limited and coarse in the most time. Such EM data is not enough to study real target’s characteristics for detection and recognition. In order to expand the data acquisition ability and get more detailed data, we propose to apply boosting decision tree method for EM scattering data interpolation. The first step is making pre-process of EM scattering data. The data includes radar cross section (RCS) under different frequency, elevation and azimuth, and needs to filter the noise in measurement system. The data should be randomly divided into training and testing datasets. The next step is the decision tree ensemble design. Using the training dataset, objective function optimization and tree parameter estimation are conducted to build EM scattering characteristic based gradient boosted tree model. The last step is to validate the model using the testing dataset. The model can be used for EM scattering data interpolation to get more completed and detailed data. Meanwhile, we performed the interpolation experiment based on the method. Numerical examples have verified the functionalities of such method in EM scattering data completion and refinement, and it will support the EM study of targets with unknown model parameters.