Resolution independent radar target recognition

The value of Through the Wall Radar Imaging (TWRI) data collected with one system is limited when training the automatic target recognition classifier of a second system because of variation in the Point Spread Function (PSF). The target image is a function of both the scene reflectivity and the PSF, which is in turn a function of the imaging radar´s parameters. Targets appears different when imaged by different systems, as a result of the PSF variation, making it difficult to create Automatic Target Recognition (ATR) training datasets including examples measured with different systems. This paper proposes a method to convert the PSF in simulated TWRIs to make them appear as if they were collected using a different radar, and so allow the sharing of data for radar ATR. The method is based on a process of deconvolution and reconvolution. The resulting images are used to test a naive Bayesian classifier and it is shown that high correct-classification rates are obtained after PSF conversion.