The Study Of Military Affairs Target Recognition And Tracking Method Based On Wavelet Analysis

Photoelectricity theodolite target tracking system is mostly used in tracking and measure of the outer trajectory, which can complete flight tracking or trajectory parameter of plane and aviation bomb. As measured object is small and dim, low contrast, the detection of puniness targets in long distance is a problem of crucial importance, because it can improve system sensibility to intruding targets and provide enough response time. Dim target recognition and tracking have important value whether in theory or in apply, which is a crucial step in infrared imaging guidance. Hence the small target recognition and tracking of long-range, low SNR (signal to noise ratio), strong clutter wave become a popular and difficult subject. In the paper, by the thorough and systemly study of military affairs target recognition and tracking method, which are solved with the realization in target automatic tracking of shooting range facility TV tracking system currently.Basing on analysis and correlative theory of wavelet transform, firstly studying the method of target detection and edge detection based on wavelet transform in the paper. In order to improve efficiency of target edge detection, studying pretreatment method of image enhancement and denoising. In frist, the recognize problem of (point, dim) moving target are researched under the sky background in the paper. The detection of weak targets in the all kinds of background, adopting Otsu segmentation method, and two different methods are put forward. The small targets pretreatment use of the median filtering and Otsu segmentation in simple background; the large target pretreatment use of the adaptive threshold, Log filtering and Otsu segmentation, the target can be extracted. Adaptive threshold used to enhance the image so that the gray background of the image becomes uniformly; the Log filter for the high-pass can remove the background effectively; the variance between the largest category is a non-classical parameters, unsupervised adaptive threshold selection method, After threshold segmentation of the image, the image will become a binary image which may contain the target. The simulation experiments show that the algorithm can effectively remove the background and has the small amount of computing advantages, which can inspect targets very well. Base on theory of the wavelet transform and multi-scale analysis, analyzing image local gray grey level, in all kinds analysise of the wavelet transform, biorthogonal wavelet algorithm is proposed to deal with dim object, as good orthogonality and symmetry, realizing optimal filter and inspection are obtained, through this way, dim object can be found.The conventional method of edge detection is difficult to attain satisfying effect between precision and anti-noise, and it is relatively sensitivety to noise. Two methods are brought forward in the edge detection algorithm of the weak target in the paper, through the studies of the morphological, after median and Log filter, a series of image preprocessing, using the edge detection algorithm of the expansion and corrosion can effectively detect the edge of the target. This method has the simple, better detection of the edge characters. Using the wavelet edge detection research through the wavelet transform, seven different wavelet filter and different threshold are used for edge detection. The experiment proved that adoptting of biorthogonal wavelet can gain the better result, and odd pels of image can be realized.Particle filter realize recursive Bayesian filter via Monte Carlo simulation. The method is suitable for any non-linear system that could be represented with state model. The theory and method of Particle filter are studied in the paper, through simulate experiment validate its capability excel to kalman tracking. A method of biorthogonal wavelet edge extraction combining Particle filter is presented, and making up its frame of tracking. Through selecting state equation of system and particle number, which realize dot target in simple background and target in shelter complicated tracking used algorithm. At last through exprement the factor of affecting the tracking accuracy are analyzed. Experiment validate, the tracker with combining robust wavelet inspect method and having many apex descriptive particle filter algorithm, which can realize steady target tracking in moving target exist local shelter.Combing the basic theory of multi-scale Gabor and BP neural net, in the paper parameter Optimizing Gabor filter and BP neural net recognize algorithm is designed. Based on Gabor better directional trait, at frist orient parameter, then searching optimal parameter of single Gabor filter in each aspecial direction, PSO optimal algorithm get Gabor filter team, which is close to optimal in the capability. In the son image distill 48 texture feature are extracted, adopting PCA (Principle Component Analysis) dimension reduction dispose, which can solve bottle-neck problem in the apply of Gabor filter, namely Gabor feature vector dimension is relatively high, as well as large computational complexity. when the algorithm is apllied to target, not only improve recognition precision, but also overcame the bug of BP algorithm get in minimum. By testing, the algorithm has better veracity and robust, and inspecting rate attain to 96.3%.For the color distortion problem of the natural color multispectral images are fused with its panchromatic images, a new fusion approach is putted forward, which integrates the advantages of both the IHS and the wavelet techniques. In the experiment, multi-wavelet is adopted, when select the high-resolution coefficients, selectting the image that have the largest variance sum of the same level high-resolution coefficients, and select its high-resolution coefficients to be the high-resolution coefficients of the fusion image. Calculate the correlation coefficients between the fused bands and their corresponding original multispectral bands, to contrast the fusion results. It can be found that use db4 and Weight Coefficients approach in the algorithm, the fusion result is the best.