采用SIFT特征的高光谱数据自动几何精校正

Automatic geometric precision correction of hyperspectral data based on SIFT feature

针对几何精校正过程中人工选取控制点误差大、未考虑高光谱数据光谱特征一致性等问题,提出了基于SIFT特征的自动几何精校正方法。首先提取图像的SIFT特征,利用高光谱数据的地理坐标定位进行局部特征匹配,然后为了进一步提取高精度、分布均匀的控制点,提出了一种分区域的随机采样一致(Random Sample Consensus,RANSAC)算法。利用航空高光谱成像仪Hymap获取的新疆东天山数据进行算法性能的分析与验证,并采用CE90/CE95以及均方根误差等指标进行定位精度的评价,提出的基于SIFT特征的自动几何精校正方法能够达到0.8像元的定位精度,并且校正前后光谱的光谱角小于0.01 rad。

Duo to including the ground control points that choosed by manual geometric precision correction were not precise, and the existing methods ignorded the spectrum consistency of hyperspectral data, an automatic geometric precision correction method based on SIFT feature was proposed to solve the problems. SIFT feature was extracted from the image and the geographic coordinate of the hyperspectral data was used to accomplish local feature matching. In order to extract high-precision and uniformly distributed ground control points, a sub-regional Random Sample Consensus （RANSAC） algorithm was proposed. The airborne hyperspectral data collected by HyMap in Dongtianshang, Xinjiang Autonomous Region, was used to analyze and validate the performance of the algorithm. The CE90/CE95 and root mean square error were calculated to evaluate the geopositional accuracy. The results show that the automatic geometric correction method based on SIFT feature can achieve 0.8 pixel geopositional accuracy, and the spectrum of the spectrum angle between warp image and corrected image is less than 0.01 radian.