Research On Feature Extraction And Analysis For SAR Images Targets Of Complex Background

In remote sensing images, the ground target detection and feature extraction are of great research value for automatic target recognition. With the development of remote sensing technology, the earth observation of the synthetic aperture radar has broad applications in military reconnaissance and identification, civilian ground facilities, surface resource monitoring. Synthetic aperture radar can provide highresolution image information, with constantly improving the resolution, so that vehicles, ships, tanks, buildings and other man-made objects have richer texture, structure, form information, more complex scattering properties and multiple features information. While speckle noise and relatively complex background information directly affect the interpretation and identification index of the SAR image target, we can improve the recognition rate and efficiency by multi-feature extraction transformation methods. The technical issue increasingly becomes a hot issue in the field of the remote sensing target recognition at home and abroad, which enhances the speed and accuracy of automatic target recognition by getting and analyzing the characteristics of the interested object in the SAR image. In this paper, based on the demand of this subject, the algorithms about the speckle noise suppression, the target detection and the target feature extraction are studied.First of all, the mechanism of SAR imaging and speckle noise generation is analyzed. And on the basis of these theories, the algorithms of reducing SAR image speckle noise are studied so as to lessen the loss of image quality information.Then, for the relatively complex background information of SAR images, the target detection algorithms are analyzed based on the threshold, mathematical morphology and pulse coupled neural network. The simulation would be done to detect target of interest. And it would like to analyze the advantages and disadvantages of the methods. At the same time, the algorithm about modifying pulse coupled neural network is proposed.Last but not the least, the methods of the target feature extraction are studied systematically. The features of Geometric and Hu moments are extracted, and the algorithm of extracting the feature of Zernike moments is proposed. Based on Zernike moments, the experiment of classifying the targets is done. And the edge information of the target is achieved by rotational invariance of Zernike moments. Meanwhile, the method of corner feature extraction is analyzed. The rotation angle estimation algorithm based on the corner information is proposed, which would improve the efficiency of the recognition in the SAR-ATR. Besides, based on the algorithm of modifying pulse coupled neural network, the method is proposed to extract the shadow information of the target, and also the characteristics of the shadow are extracted. In the paper, on the one hand, through a large number of experiments using the MSTAR database, the performance of the above feature extraction methods is verified. And on the other hand, the target feature extraction software is designed and completed, which provides the strong data support for the target classification.