Study On The Technology Of Spatial-temporal Non-stationary Heavy Clutter Suppression And Dim Small Moving Target Detection

The widespread applications in more and more fields have generated a great deal of interests for detecting and tracking dim small moving targets with heavy clutter interference in image sequences. Because the interested targets are so small inherently, and usually immersed in strong background clutter, it is very difficult to detect and track them. And this dissertation concentrates in the two key technologies of background image clutter suppression and targets detection in additive quasi-WGN.Firstly, this paper proposes the system of detecting dim small moving targets in image sequences based on suppressing heavy clutter distinctly. That is, a two-level policy is adopted that suppressing image clutter firstly and then detecting targets with matched filter. And the tasks of clutter suppression are summarized as two points: one is removing background energy and improving SCR, the other is transforming the complex, unknown, spatial-temporal non-stationary distribution of original image to simple, known one. The evaluation technology of clutter suppression is developed, and four engineering indices are suggested: tests of normal property, whitening character, LSCR gains and detecting performance of single frame.Secondly the paper discusses how to make definition of the dim small target in images, and the analysis has been done in terms of human vision and machine detection respectively. The definition of local signal to clutter ratio (LSCR) and its reason are expatiated. The image target LSCR demarcation of dim character for human vision is obtained by engineering experiments. Using Neyman Pearson criteria the relevant demarcation for machine detection is given in terms of detection and false alarm probability.Thirdly the technology of adaptive clutter estimation and suppression based on quasi-stationary sub-image partition is proposed and its detailed computer simulations are completed. To overcome the non-stationarity in whole image spatial domain, is presented a method of partitioning original image to some quasi-stationary sub-block in which adaptive filter is adopted for predicting and suppressing clutter. And two methods of partition are proposed: quad tree moment segmentation and FCM cluster segmentation. The raw image data processed by pre-smoothing as input data of adaptive filter can diminish background prediction distortion due to target gray-level spread.Fourthly, an effective morphological neural network of background clutter prediction for detecting dim small targets in image data is proposed. The target of interest is assumed to have a very small spatial spread, and is obscured by heavy background clutter. The clutter is predicted exactly by morphological neural networks and subtracted from the input signal, leaving components of the target signal in the residual. The traditional 3-layer feed forward BP network modal of morphological opening and closing operation is modified by extending the input layer data. For tracking complex background including different sub-structure, the raw image data is partitioned to some sub-block, in which the training samples are chosen for optimizing the weights of structuring element in the corresponding block, which leads to better clutter estimation.Fifthly, a fast algorithm of background clutter estimation and suppression based on spatial-temporal adaptive clutter classification is suggested. If global motion in image sequences caused by camera doesn't exist or the motion is estimated and compensated, the image pixels are classified, by learning temporal gray-level moment of input sequence, to two categories: temporal stationary clutter and not. The stationary clutter is caused by stilled background with noise, and the non-stationary one results from different factors such as background fluctuation or variation, camera motion and targets existence. And a temporal filter and a spatial adaptive filter are applied to the two for suppressing clutter respectively. It combines the advantages of temporal and spatial filter, which is a fast, precise and adaptive method with low storage price.At last, a spatial-temporal integration system for detecting dim small moving targets in WGN is proposed, which is in favor of implementing in real time, and detailed performance analysis is discussed. According to the priori information or experience, the target energy is integrating in spatial domain (single frame) firstly. Then by binaries processing, the better probability of detection and false alarm are obtained. Then we make use of the movement continuity, and the target energy is integrated further in temporal domain. For the spatial integration detection, the performance of linear and nonlinear combination is analyzed in detail. For the temporal integration detection, the performance of three-dimensional trajectory integration and temporal projection integration are both discussed.PC simulation and its results analysis are carried on with all techniques discussed. Experiment results show that all the research works in this paper have important values in the theory and application on detecting and tracking dim moving targets in image sequences.