Small Target Detection Based On Higher-Order Cumulant And Wavelet Energy Transform

Small target detection is a very important content of image processing. Small target detection has wide engineering application such as military affairs, communication, weather predicting, GPS, land and ocean exploring, seismic and biomedical signal processing etc. Especially, small target detection is a very important content of military affairs, such as airborne early warning, maritime surveillance and so on. In recent years, because of infrared and visible light reconnaissance, surveillance equipments' application is a large number of in the military, and also promote the multi-target detection technology's development rapidly.In this paper, higher order cumulants and a variety of outstanding modern digital signal processing methods are combined, which is used in the multi-dim-small target detection with the Gaussian noise and natural context.Digital image can be regarded as consisting of three components: the target image, the background image, the noise image.The so-called small target is that when the relative position of the imaging systems and the target is more distant, maybe the target's diameter is a few or even a few ten meters, but it displays one or a few pixels area at the performance of imaging plane, while,it usually also known as the point target(PointTarget).In order to detect the low-intensity exercise of small targets from two-dimensional images, the signal processing in space filtering method is used in this paper. According to the identity of higher-order cumulants which has been applied in other fields, for the spatial characteristics of infrared image small target, the feasibility that in theory AWGN can be completely filtered with the good inhibitory effect of higher-order cumulants of Gaussian noise is derived in the two-dimensional image. And it is verified its accuracy by matlab simulation, as well as the scope of application. Compared with other existing filters, the adaptive filter based on higher-order cumulants that introducted in this paper successfully extracted all the small targets, and retained the characteristics of natural texture. While, there is not only the loss of information, but serious damage to the background in the two-dimensional winer filter and two-dimensional extremal median filter. It is difficult to discrimination of the extraction of target after the two filters' filtering, in particular, submerged in the background of the objectives. The targets drowned in background is very difficult to be segmented in projects, and these disappeared small targets will cause irrecoverable loss in military.It contains small targets and natural backgrounds after the higher-order cumulants adaptive filter. On this basis, the method, which consists of the wavelet energy transformation, reconstruct, data integration and threshold segmentation, is going to be used in the background segmentation and the target image. The method introduced in the paper's fifth chapter , can segment unnatural small targets out for military valid reference. Target image is only some pixel grayscale singular point. Because of its small share, the lack of size, shape, texture and other structural information, the only available information is the target strength. Background image usually has a "strong correlation" characteristics, so it occupies all low-frequency space in the entire scene image. At the same time, because of the inhomogenous distribution between hot scenes and sensor internal, the background image is a non-stationary process. The local gray value changes may be larger in the image, and it shows strong "up and downs" characteristics. In addition, the background image also contains some high frequency components of the space domain, which are mainly distributed in the edges of every homogeneous area in the background image, such as the junction line of the sea and the sky in background image.At the same time, wavelet transform is used in image decomposition, image reconstruct, image fusion etc. to make the high-frequency-channel images with them. At last, we get all the high-frequency information the small targets from the image. The small targets are entirely extracted from the background and the noise by the method of adaptive threshold segmentation. From the experimental data, small targets detection shows more advantage by the method that consists of higher-order cumulants adaptive filter, wavelet energy transformation, data fusion and adaptive threshold segmentation.That image processing developing rapidly nowadays make great achievements. Small target detection in the military is researched further and further, as well as its application. It will appear more and more new demands in the field, so there is a huge space for development.Therefore, a image with AWGN filtered by higher-order cumulants adaptive filter that introduced in this paper, which can segment small targets information successfully and retain the bulk of the natural background texture. In this way, target detection algorithm with natural texture background based on wavelet transform is introduced in this paper. The method specific to the natural texture (non-obvious features) the background is about wavelet energy, image reconstruct, data integration and threshold segmentation. Compared with the traditional segmentation methods, this algorithm can filter background similar to targets' gray-scale sucessfully, while retaining the small targets.With image processing developing and the theory suggested from related fields improving, that multi-small-target detection is bound to apply and promote with greater depth and breadth in various fields, will promote researchers more in-depth study on multi-small-target detection.