资源三号卫星影像无控制区域网平差

Method and GCP-independent block adjustment for ZY-3 satellite images

国产高分辨率光学卫星影像的无地面控制精准定位是实现境外大范围区域或大规模遥感应用和中小比例尺测图的前提。针对资源三号卫星(ZY-3)影像的几何特点,以航空摄影测量中约束问题最优化方法"交替趋近法"和基于RFM的最小二乘平差为基础,提出了一种易于并行化、高效的高分辨率光学卫星影像无控制联合区域网平差方法GISIBA(GCP-Independent Satellite Imagery Block Adjustment),通过构建"平均值"虚拟控制点来解决无控制区域网平差中的"秩亏"问题,便于分析无控制区域网平差结果与影像数据的覆盖次数及时相之间的关系。首先,交替趋近法被用来实现并行处理平台基础上的待平差未知数初值的解算、中等尺度以上粗差的自动检测与剔除,并根据计算结果赋予所有未知数一定的先验权值;然后,通过最小二乘法实现大型法方程矩阵的解算,获得满足高精度影像产品生产制作需求的高精度定向参数;最后,利用多组典型区域的ZY-3影像数据试验验证了该方法的精度和实际性能。结果表明,无控制区域网平差达到了优于6.0 m/5.0 m的平面/高程精度,所提方法为全球地理信息资源建设工程、国产光学卫星影像高精度影像产品生产提供了技术保障。

The spatial and radiation resolutions of domestic satellites have been remarkably improved in recent years. The sensor design and geometric calibration technologies have improved with the development and application of stereo mapping satellites(ZY-3). However, the imaging systems typically use the linear CCD imaging technology. The long focal length and narrow field angle causes the geometrical model to have 3 D parallel projection characteristics, and the satellite images will still have residual system errors caused by the drift error of satellite-borne GPS/IMU and asynchrony between the pose and trace of the satellite, especially after the rigorous on-orbit geometry calibration. A block adjustment technique is still required to meet the requirements of remote sensing monitoring and mapping application. Therefore, satellite images-based accurate positioning without ground control point information is the precondition to obtain the global geographic and resource environmental information and monitoring changes in the global resource environment. The accuracy of "ZY-3" satellite image is improved to 15 m after calibration and its internal precision is better than 1 pixel. After the overall block adjustment without GCPs,the plane and elevation accuracy of the image can be improved to 5 m(medium error). In this study, on the basis of the widely used optimization method of solving the constraint problem in photogrammetry–alternating direction method(ADM) and RFM least-squares block adjustment, we propose a GCP-independent block adjustment method for large-scale domestic high-resolution optical satellite images–GCP-independent satellite imagery block adjustment(GISIBA) based on the geometric features of ZY-3 satellite images. The proposed method is highly efficient and easy to parallelize.GISIBA of satellite images can be considered as the overall block adjustment of the multi-source optical satellite imagery with specific constraints(distance, angle, etc.) when no ground control points are available. Most GCP-independent adjustments use a form of virtual control points. However, the precision of these virtual control points is low(varied system error), and the precision is inconsistent in the measurement area. Thus, the GCP-independent adjustment is a type of block adjustment under different precision controls. The law of error propagation of this approach is complex, and gross error detection and positioning are difficult to perform using this approach. This study presents an "average" virtual control point-based stereoscopic GCP-independent block adjustment method for large-scale satellite image GISIBA. On the basis of the automatic and reliable acquisition of uniformly distributed image tie points, the method comprehensively uses the "ADM" introduced from aerial photogrammetry and RFM-based least squares adjustment algorithm to realize the combined block adjustment of satellite images. First, the "ADM" is used to solve the initial values of the unknowns and to perform automatic detection and elimination of the above medium-scale gross error based on the parallel processing platform. All unknowns are assigned priori weights based on the results. Next, the RFM-based least square method is used to solve the large-scale reformation normal equation to obtain the orientation parameters with high-precision, which meets the production requirement of high-precision image products. Block adjustment by constructing virtual "average" control points addresses the "rank" problems in the GCP-independent adjustment and improves the state of normal equation of the block adjustment system, which benefits the stability and fast convergence of the block. Moreover, the method makes it convenient to analyze the relationship among the data coverage, imaging time interval, and satellite image GCP-independent block adjustment. In addition, parallel processing based on the OMP parallel method is used to realize the parallel processing of the "ADM" and multithread parallel computing based on least-squares adjustment to ensure the efficiency of block adjustment.We used multiple sets of the ZY-3 satellite image data in typical regions to verify our method. The following experiment results are summarized as follows: 1) On the basis of the widely used optimization method of constraint problem called the "ADM" and RFM leastsquares block adjustment, the proposed GISIBA method is easy to parallelize and is highly efficient in terms of reliability, accuracy, and performance of the developed procedure. 2) In this method, virtual "average" control points are built to solve the rank defect problem and qualitative and quantitative analyses in block adjustment without control. Assuming the positioning accuracy is located on the same number order(such as 50 m), the final positioning accuracy of satellite image must be improved after GCP-independent block adjustment by using the virtual "average" control points. The final positional accuracy is stronger than the worst initial positioning accuracy of the original image.Furthermore, the increase on the coverage of satellite images does not consistently improve the overall positioning accuracy. However, the use of considerable high-resolution satellite images to cover the same area improves the positioning accuracy after the final block adjustment in the statistical sense. The horizontal and vertical accuracies of multi-covered and multi-temporal satellite images are greater than 6 m and 5 m, respectively. 3) The mosaic problem of adjacent areas in large area DOM production can be solved when third-party geographic information data are introduced as horizontal and vertical constraints. This approach is considered as weak-sense auxiliary control in the block adjustment process.