Unknown Orbital Parameters Of Remote Sensing Images, Geometric Correction Model

In the high-resolution satellite remote sensing image, due to some inevitable factors, such as the change of satellite platform's posture, height, speed, the earth rotation,and so on,the images obtaind will produce different degree of geometric distortion,which is relative to the real earth's surface.Before using these remote sensing data for other kinds of field,we should eliminate these aberrations to accurate positioning.According to the satellites attitude and platform parameters in imagery moment, the strict physics imagery model used to correct distortion of remote sensing images is based on collinearity condition equations which can describe a rigorous geometry relationship between point coordinates and target ground object coordinates. But because of unknown orbital parameters, large flightaltitude and small view angle, space resection of single image is difficult to obtain satisfactory results.In order to overcome these difficulties, many generalized geometry correction model which does not involve satellite orbital parameters was proposed and obtains widespread concern. They can effectively correct geometric distortion in remote sensing images and can achieve high correction precision, so often used in high-resolution remote sensing satellite preprocessing.Depend on the imagery theory of linear array push sweep sensor launch on the high resolution satellite, this article digs into the creation process and computing method of strict physics imaging model, analyzes difficulties in space resection of single image settlement.Then, from two aspects include model establishment and solving, the paper emphatically studied affine transformation model, direct linear transformation model, a rational function model as generalized geometric correction model. Aiming at the normal equation ill-posed problems generally exists in model solution, analysis produces pathological reason, diagnosis pathological degree,and came up with ridge estimators and two-step method to improve the morbid state geometric correction models. Finally the experiments on different models and methods overcoming with ill-posed state all get the desired results. After the ill-posedness of three models, we found that the solving process ill-posedness of affine transformation model which is better accorded with strict imaging model is the most serious, the second one is rational function model, and the ill-posedness of direct linear transformation model was the lightest.The results of three models'experimental show that ridge estimate and two-step method all can effectively overcome model's ill-posedness. But the ridge parameters in ridge estimate are difficult to determine, and two-step method can intuitively determine the regularization parameters by using L curve method.Analyzing show that point correction precision of the rational function model improved by the two-step method can reach subpixel accuracy, which has achieved the purpose of eliminating image geometric distortion, and this is an effective generalized geometry correction model.