针对环境与灾害监测预报小卫星(HJ)图像存在较大整体几何误差,且无规律可循,难以使用全局模型模拟整景图像几何变形,而基于手动方式选取控制点的全局模型和局部模型都不适合于HJ图像几何精纠正的问题,提出一种基于加速分段测试特征(features from accelerated segment test,FAST)算法测点并用局部模型进行几何精纠正的优化方法。首先以FAST算法获取大量地面控制点(ground control point,GCP);再使用多项式模型对GCP的均方根误差阈值、潜在不匹配和实际不匹配GCP数量进行相关分析,据此修正FAST参数,筛查GCP误点;最后使用局部模型完成几何精纠正。此外,使用散点图和空间插值等方法建立适合于HJ图像几何精纠正结果的评价指标。检验结果表明,该方法能使纠正误差控制在1.5个像元内,纠正后的图像能满足中分辨率尺度的应用要求。
Owing to relatively serious irregular geometric deformation of the whole image, it is difficult to simulate the deformation property of the whole large HJ satellite image by using the global model. Neither the global nor the local calibration model based on traditional manual selection of the ground control point ( GCP ) is suitable for geometric accurate rectification of the large HJ satellite image, and hence the authors proposed an optimization local model of geometric rectification method for calibrating HJ satellite remote sensing image after selecting GCPs with the algorithm of features from accelerated segment test(FAST). This method can be described as follows: First, plenty of GCPs are automatically obtained based on FAST algorithm; Then, a polynomial model is employed to analyze and evaluate the points' matching ratio between non -matching GCPs and actual GCPs in the RMSE threshold with the correlation analysis method;In addition, the total error GCPs' numbers are calculated according to the analysis of the results and the choice of appropriate root-mean-square error threshold to eliminate error GCPs;Finally, the rectification is completed with local model of Linear Rubber Sheeting. Besides, the authors have established a criteria system which is suitable for the registration result by evaluating HJ satellite remote sensing image, including residuals scatter plot, spatial interpolation method and so on. The evaluation results show that the error of calibrated HJ satellite remote sensing image can be restricted within 1. 5 pixels, and the calibrated image meets the requirements of mid-resolution's application satisfactorily.