Research On Target Robust Detection Based On Random Perturbation Model And Subspace Model

The task of target detection is to search targets in the region of interest based on the known target feature vector(the vector describing target features),so the reliability of target feature vectors is crucial.In fact,the known target feature vector is often unreliable,i.e.,there is a difference between the known target feature vector and the real target feature vector.Under such mismatched situations,traditional detection algorithms designed for matched situations suffer severe performance losses.To solve this problem,this paper considers how to design detectors that are robust to the mismatched target feature vector.The main contents are as follows:1.Radar distributed target robust detection.In recent years,with the development of radar technology,the resolution of radar has been continuously improved,which makes the target echoes often occupy several range cells(the so-called distributed or range-spread target).To make the H1 hypothesis(the hypothesis that target exists)more plausible in the mismatched case,the received signal under the H1 hypothesis is modeled as the sum of noise,useful target echoes and fi ctitious signals.Two adaptive detectors are designed according to the Rao test and Wald test.The proposed detectors are proven to possess constant false alarm rate properties against the noise covariance matrix.Experiments show that the proposed detectors have strong robustness against the target feature vector mismatches.2.Robust detection of hyperspectral subpixel targets.The so-called subpixel target refers to a target which only occupies a part of a pixel due to the low spatial resolution of hyperspectral sensors.This paper examines the problem of detecting subpixel targets in hyperspectral images.Considering that the subpixel target feature vector is not always reliable(e.g.,due to spectral variability),an interaction subspace model is designed to deal with this problem.In this subspace model,the second-order interaction terms are introduced to better describe the spectral variability,thereby improving the robustness.Based on this subspace model,two detectors are derived according to the one-step generalized likelihood ratio test and its two-step variant.Experiments show that the robustness of the detectors is significantly improved after using the subspace model and the proposed two-step detector exhibits good robustness in mismatched cases.