Algorithm Fusion Approach to IRST Signal Processing (Ill): Nongaussian-Corrected Filter Fusion

摘要

The reduction of background clutter in Infrared Search and Track (IRST) systems remains a challenging problem. False alarm inducing clutter is typically nonstationary and nongaussian and may not be adequately treated by linear matched filter techniques. Previously, we examined the simultaneous use of two filters, 'filter fusion', to reduce the false alarms. This method depends upon the assumption that the residuals of two sufficiently different filters will be uncorrelated. In the present work we make explicit use of the nongaussian character of the clutter, using a threshold that depends on the local nongaussian character. The introduction of a nongaussian correction often improves the performance of both linear and nonlinear filters, including the optimal linear filter constructed for the scene. Fusing the nongaussian adjusted filters gives a further increase in performance. The use of the nongaussian correction appears to increase the robustness of the filters and the filter fusion, i.e., the performance from frame to frame within a given scene and from scene to scene is consistently improved and made more regular.