Study Of Empirical Mode Decomposition And Its Applications To Infrared Target Detection

The infrared target detection is one of the key technologies in the infrared guidancesystem, which can provide the target’s shape and location information. How toimprove the detection probability of the small target in low SNR and how to getcontinuous and sharp edge of the big target are two meaningful and challenging majorresearch topics.The empirical mode decomposition method is a new multi-scale and fullydata-driven signal processing tools which can especially manage non-stationary andnon-linear signal. EMD has been widely used in one dimensional signal processing,but extended to the image processing, EMD method and its applications still havemany issues to be resolved.This thesis conducts a study on EMD and its applications in the infrared targetdetection under complex background. Firstly, on the basis of the EMD theory, thispaper proposes a kernel regression interpolation based BEMD algorithm, which get thefitting surface by conducting a non-parametric kernel regression matrix to interpolatethe scattered extreme points. The experimental results show that the proposedalgorithm can significantly reduce the time complexity, avoid boundary effects andimprove the decomposition quality, and thus make the resulting intrinsic modefunctions more meaningful. Secondly, a new infrared small target detection algorithmunder the complex cloud-sky background is presented which combines the EMD basedbackground suppression method and the variable weighted pipeline filter method. Theintrinsic mode functions extracted by EMD which represents the image’s highfrequency information such as target, clutter and intensive cloud edges is considered tobe the background supresseion image, and then the target’s trajectory can bedetermined by variable weighted pipeline filter. The experiments show that this methodcan make full use of target’s space and time dynamic information, and realize the smalltarget detection under low SNR through effective target signal energy accumulation.Finally, a new infrared edge detection method based on EMD and bidimensionalspectral features theory is proposed. The first instrinsic mode function achived byEMD uses the quaternion analysis to get its spectral features, and after set theappropriate threshold, we get the edge detection image. Experimental analysis showsthat the infrared target’s edge got by this method is sharper, more continuous, and morein line with the nature of the image characteristics.