Research On HJ Satellite Multi-spectral Camera CCD Band-to-Band Registration's Accuracy Evalution

Multi-spectral camera takes images at different spectral bands, which can be used to generate thematic image products after band registration and data fusing, enriching vastly the information contained in images and improving the image's application potential as well. Currently, multi-spectral camera has become one of the most important sensors of remote sensing satellite.There are many factors that have impacts on band to band registration (BBR), such as the temperature changes, the structure stability and so on. Because of elevation changes and the inner distortion of the sensor, the images of the multi-spectral camera at different bands usually have large nonlinearity geometry distortion. Therefore, only by plane translation and rotation to realize the BBR is hardly to satisfy the products' quality. As high spatial resolution satellite, the accuracy of BBR is critical to positioning precision, classification accuracy and applicability.Based on the traditional registration methods and incorporating with the feature of HJ satellite, this paper evaluates the BBR error of 4-bands HJ CCD, due to device assembling, lunching, and transporting. In order to express the BBR errors more accurately, several BBR algorithm were proposed in this paper according to the features of HJ satellite based on registration algorithm and sub-pixel positioning method. These algorithms includes grey-core algorithm, Gaussian line fitting algorithm, Hough intersection algorithm, Gaussian curve surface algorithm, and Ground Target algorithm. These algorithm analyses the BBR errors from multiple aspects of image, including gray-scale of image, property of edge lines, the properties of partly curve surface, and the digital number distribution of research area's.Based on these algorithm, this paper calculated the BBR errors of HJ CCD in both along-track and along-scan directions. Results show that the BBR of HJ CCD level-2 product is less than 0.5 pixels and it reaches to 0.1 pixels in some products. Moreover, calculation results from all of those algorithms have steady and comparable result, which ensure the accuracy from different views and perform the cross-validation.