A Study On The Detection Of Small Targets In IR Image Sequences With Heavy Cloud Clutter

In order to detect small dim targets in IR image sequences, a temporal processing technique is investigated. At first, models for the temporal and space behaviors of the background noise, target and clutter, on a single pixel basis, are introduced. Then based on the temporal difference models, we formulate the detection problem in 2 steps, correlation detection and generalized likelihood ratio test (GLRT). After correlation detection step, noise pixels in the image sequences are almost suppressed, and only target pixels and a few clutter pixels can pass the detection threshold. In order to further the targets detection in these few pixels, an improved GLRT method is developed. This improved GLRT method can suppress the clutter pixels sequentially and enhance the performance of the temporal detection method. Theoretical analyses show that this method can detect targets on very high detection probability and very low false alarm probability. The effectiveness of the technique is demonstrated by applying it to real world infrared image sequences containing cloud clutter and airplanes flying at long range.Second, based on the analyses of early warning satellite IR targets detection application, it is believed that this application is kind of fast targets detection in slowly evolving clutter. Then a temporal targets detection approach is put forward. This approach can be thought of as a reform of the former temporal technique. Theoretical analyses and experiment research validate the effectiveness of the approach.At last, a temporal-space method to detect small dim targets in IR image sequences is developed. First, an adaptive Wiener filter is used to suppress the clutter background of one single frame, so a white noise sequence is got. Then we use a 3D targets detection pipeline to decide the locations of the targets. Experiment results show that this method is an efficient 3D targets detection method.