Accurate Positioning Of Non-cooperative Satellite Remote Sensing Image Based On Registration Technology

With the development of aerospace technology, satellite remote sensing image has played a huge role in the environmental monitoring, disaster prevention and reduction, national defence etc. But due to the influence of various factors, the original satellite remote sensing image has various kinds of geometric distortion. In order to make efficient use of the satellite remote sensing image, the first task is to establish the accurate geometric relationship between the pixel coordinates and the corresponding ground coordinates, namely accurate positioning. Non-cooperative satellite remote sensing image, due to the loss of the longitude and latitude information, does not known the accurate coordinates of the ground objects, and greatly reduced the practical value of the image.Taking the non-cooperative satellite remote sensing image as the research object, this paper proceeding with image registration technique, aims at solving the automatic accurate positioning problem of the satellite remote sensing image, so that it can play a greater role in change detection and target recognition.① Research on the automatic image matching algorithm based on local invariant features. On the base of the classification and summary of the SIFT algorithm, focuses on the ASIFT algorithm which is invariant to the perspective view changes. Considering the high complexity of ASIFT algorithm, the fast SURF operator is used. The performance of the ASIFT algorithm and the improved algorithm based on ASIFT are compared by three experiments. The results show that the algorithm in this paper has improved the execution efficiency in some degree.② According to the fact that the non-cooperative satellite image has low contrast, histogram equalization is used to enhance the image, so that the image quality can meet the requirements of registration; Considering the problem that the non-cooperative satellite image can not directly match with the Google map, the Google landmarks are mapped to a representative satellite image through manual annotation and data fitting as the motherboard, and then compared the positioning accuracy of the two kinds of fitting model including global and local. The results show that the local fitting model based on thin plate spline can better deal with the condition of topographic relief and with higher positioning accuracy; Non-cooperative satellite image has the characteristics of time continuous and large amount of data, the manual registration of every image is not practical. To solve this problem, firstly, the new satellite image will be matched with the motherboard by using the proposed algorithm in this paper; secondly, the corresponding landmark in the motherboard will be mapped to the new image, thus the longitude and latitude coordinates automatically estimate of the new image will be finished; finally, evaluation of the new positioning accuracy was made, the results show that the improved matching algorithm in this paper has a good adaptability to the cloud and snow shielding, perspective changes, rotation and translation of the non-cooperative satellite image. Positioning accuracy of the new image has reached the desired effect.③ An example of the accurate positioning system is build and programmed based on VC++, which includes the ASIFT algorithm, the algorithm in this paper, data fitting, automatic processing etc. The software automatically registers the new image, it meets the requirements of the non-cooperative satellite image automatic processing.