Research On Geometric Correctionand Object Recognition For Remotesensing Image

The rapid development of remote sensing technologies enables humans to better, faster and more comprehensively observe the world. Many objective factors, such as the altitude variation of imaging sensor, the attitude variation of carrying platform and topography, will cause geometric distortion of the image. With the constantly enhanced ability of the remote sensing for the earth observation, how to realize the high-precision acquisition of Earth spatial information by processing the remote sensing data is the basis of expressing the advantages of the remote sensing. Therefore, it is no delay to study on the high-precision geometric correction theory and method and improve the potential application of object recognition for the remote sensing data. However, as the diversified and multi-level development of remote sensing systems, the different imaging characteristics present new demands and challenges on the geometric processing technology for remote sensing image. As a result, traditional methods are no longer applicable, and new methods are needed which are expanded according to the imaging characteristics. In view of the application requirements of the effective localization and recognition for the surface features, this thesis focuses on the key techniques of geometric correction including the solution of exterior orientation elements, the extraction and optimization of control points and the construction of imaging geometric model for the different remote sensing data by different platforms in the non-ideal imaging environments. The final purpose of this thesis is improving the recognition precision of remote sensing object through implementing geometric correction to effective recovery the position and shape information of remote sensing objects.Firstly, the coordinate systems and their transform models are briefly introduced; two imaging models of remote sensing image are introduced and their application characteristics are analyzed; the distortion of remote sensing image caused by external factors are detailedly analyzed, and the nonlinear effects of image resolution caused by view angles are discussed in detail; the projection errors are derivated in theory and analyzed from the vertical and oblique imaging. These contents constitute the basis of geometric correction for the remote sensing image, and they are the theoretical foundation for the following chapters.Sencondly, rigorous geometric correction algorithm is studied for linear pushbroom satellite image with satellite imaging parameters. As the traditional method can not accurately describe the whole flight path, an iteration algorithm is proposed for calcuating exterior orientation elements every line to solve the large correction error in the flight direction. As the large imaging-yaw angle results in uneven correction error in the scanning direction, the error propagation of off-nadir point is theoretically analyzed and the improved self-adjusting polynomial is proposed to compensate the estimation deviation of the pixel coordinates. The proposed algorithm based on iterative solution and self-adjusting polynomial is applied to the real satellite remote sensing data, and the experimental results show the higher correct precision in both row and column direction than conventional algorithms. High-precision geometric correction for satellite remote sensing image is a fundamental guarantee for positioning the remote sensing objects.Thirdly, for geometric correction of the aerial remote sensing image imaging in large view angle, the feature point extraction and the geometric model estimation are mainly researched. Aiming at the conditions that the local distortions and the differences of intensities have influence on point extraction results, a feature point extraction algorithm based on multiple view space is proposed. Compared with the traditional algorithms, the number and correct ratio of matched points are effectively improved by the proposed algorithm. As the different regions have different distortions for the large view angle image, the performance of single correction model is not precise enough for correcting the whole image. According to the application conditions and characteristics of polynomial model, the piecewise polynomial estimation algorithm is proposed based on the control point distribution. Experimental results show that the proposed algorithm is more suitable for correcting the large angle image and have the higher correction precision than conventional algorithms. High-precision geometric correction for aerial remote sensing image with large view angle is an important prerequisite for the efficient recognition of remote sensing objects.Finally, based on the effective correction of the position, shape and texture of remote sensing image, object recognition technologies for the corrected remote sensing image are researched. To solve the problems of the difference of intensities between remote sensing images and the efficiency of searching object in an image, an object recognition algorithm based on phase congruency feature and chaos-particle swarm optimized searching is proposed. Experimental results show that this algorithm greatly improves the computation efficiency and ensures the correct recognition rate, and validates the effectiveness of the proposed algorithm in the third chapter. For the small object recognition by using the global invariant feature which is independent of spatial position, pose and illumination variations, considering that multiscale autoconvolution feature(MSA), which has the prominent comprehensive performance, is very sensitive to illumination change, a new algorithm of extracting affine invariant feature is proposed based on the MSA combining with texture structure analysis. The experimental results indicate that the new feature has better adaptability in various environmental conditions and the recognition accuracy of it is superior to MSA and other improved methods. For the small object recognition with the new feture for the before-and after-correction images in the fourth chapter, the experiments validate that the geometric correction is an important procedure to realize the efficient recognition for the large view angle image. In addition, two indexes for evaluating the information loss are proposed, which provide a good way to describe its effect on object recognition results.