Target Detection And Classification Based On Random Observation Vector

Compressed Sensing (CS) is a new direction between mathematics and information science. Based on the application of Compressive Sensing for image reconstruction, this paper explores the application of Compressive Sensing for target detection, SAR image classification, hyperspectral image classification. Details are as follows:(1)Present a method for target detection based on compressive sensing, the image is natural image. We obtain the target’s probability density function by a mixture of factor analyzer (MFA), and add probability density function of the target, as priori knowledge of the target.So the target is reconstructed more clearly relative to the scene elsewhere and the target is prominent in the reconstructed image.(2)Use the fact that each class of the targets of MSTAR dataset forms a non-linear, smooth manifold and the random projection maintains the structure of the manifold with high probability, we present a new method for the MSTAR SAR image target classification.(3)According to the characteristics of hyperspectral image data:high dimension and have correlation between bands, we proposed the method for the spectral dimensionality reduction by random observation, and then classified the random vector of the image by supervised classification method.