Chinese Journal of Computers   Full Text
Title	Adaptive Spatial-Temporal-Spectral Filters for Point Target Detection
Authors	HU Jing1) ZHANG Tian-Xu2) LIN Liang-Kui3)
Address	1)(State Key Laboratory for Image Processing and Intelligent Control, Huazhong University of Science and Technology, Wuhan 430074) 2)(Institute for Pattern Recognition and Artificial Intelligence, Huazhong University of Science and Technology, Wuhan 430074) 3)(School of Electronic Science and Engineering, National University of Defense Technology, Changsha 410073)
Year	2008
Issue	No.1(161—169)
Abstract &   Background	Abstract In allusion to the problem of dim moving spot target detection in the presence of highly structured background clutter, the authors put forward a new algorithm of adaptive spatial-temporal-spectral multidimensional filter (called ASTS filter) according to the applicable principle of the Wiener-Hopf equations, and present a new algorithm to implement multidimensional information fusion. In this way, there are three processes. Firstly, the quality of inosculate of spatial-temporal-spectral multidimensional information is studied. Secondly, the filter template which in allusion to the moving dim spot target is constructed. Lastly, the Recursive Least Squares (RLS) filter is applied to the multi-spectral infrared images with highly structured background clutter. In general, the gray of pixels will be used as the input parameter of the RLS filter, and the large-size matrix will bring the huge computational load. The spatial-temporal-spectral multidimensional information is used as the input parameter in the new filter, and then it can shun the large-size matrix and reduce the computational load. By the way, the new filter can get the higher output SNR (Signal Noise Ratio). A great deal of experimental data testifies the well performance of this algorithm in filtering. keywords multidimensional information fusion; multidimensional adaptive filter; recursive least squares; point-source target detection; adaptive spatial-temporal-spectral filter (ASTS filter) Background This research is supported by the National Science Foundation of China under the grant No.60135020 and PRASI under the grant No.01-03. The title of the project is The Theory and Key Techniques of Automated Recognition of Targets under Complex Environments. It is a hard problem for the detection of dim moving spot target under the highly structured background clutter. And the fusion of multi-spectral images information is also a question for discussion currently. Many methods for the fusion of multi-spectral images information to detection and recognition of dim moving spot target have been proposed. However, because of the highly structured background clutter is a hard block for detect moving spot target problem, the methods available often yield poor results. In addition, these methods usually suffer from huge computation load that is problematic in real-time systems. The research group has been working the detection of dim moving spot target of multi-spectral images since 2000. In this paper, authors propose a new spatial-temporal-spectral multidimensional information fusion algorithm based on RLS, which has the potential to reduce the computational complexity and raise the ability of resisting noise and highly structured background clutter.