Geometric Correction Of PHI Hyperspectral Remotesensing Image Bases On POS Synchronization

Geometric correction of Airborne Hyperspectral remote sensing data is always important in hyperspectral image preprocessing. Based on the traditional algorithm, it requires a huge amount of computation necessary search or other large amount of calculation, so it will spend a lot of time. In order to improve the processing speed, this thesis is aimed at the geometric correction part of the optimal resampling algorithm.POS/AV position and orientation system of Applanix company is used in POS system of airborne sensor, not only its position accuracy is up to5cm-30cm, but also its direction accuracy is up to20s-30s.So it can provide us accurate position and attitude information. This article is to take full advantage of the matching position and attitude data, such as the high-precision parameters of latitude, longitude, altitude, roll angle and pitch angle, using geometric correction algorithm resample each pixel gray value and the actual coordinates ground element, thus eliminating geometric distortion remote caused by the sensing platforms.This thesis first introduces the features and applications of hyperspectral remote sensing, the reason of geometric distortion and the requirement of parallel in geometric correction. The largest impact for Geometric distortion is the airborne platform instability, so geting high-precision position and attitude data is particularly important.For the reason of geometric distortion, it introduces the working principle of the four components of POS systems in the second chapter. And it also explains particularlly the reason of GPS/IMU can get high accuracy position and orientation data. Before the algorithm of geometric correction, line deletion and line repeat the error appear in the hyperspectral image. After the Matlab simulation analysis, we find the law of the line number.the realization of the image restoration, rehabilitation after a good overall result. In addition to sensor errors as one of the reasons for this internal distortion, line spectral information transmission is also a major factor limition after analysis. Then we realized the image restoration, the recovery of overall effect is very good. Then it makes a detailed presentation for the coordinate transformation and resampling in geometric correction algorithms. In traditional the method of resampling, this thesis presents a new resampling algorithm called local search inverse distance method, this algorithm does not require a lot of computing and searching. After this algorithm processing, the overall effect of the image is also very good. Finally, in traditional the features of hyperspectral remote sensing data to single-band and characteristics of multi-core, it proposes a a multi-threaded parallel programs. In the process of coordinate transformation, the data reads frequently, but the data reading is single-threaded, so the data parallel in the coordinate conversion section is not feasible. In the same time,the resampling part calculates as the number of bands and rows,so it is suitable for multi-threaded parallel.In parallel computing, it proposes the parallel mode of dynamic binding cores, then makes a detailed analysis for the parallel performance. Finally, the result shows that the mode of multi-threaded binding kernel is not only improves the efficiency in resampling greatly, but also greatly reduces the memory overhead caused by the load imbalance.With the fast development of high-spectral image resolution enhancement and multi-core technology, how to take full advantage of high-performance computing to improve resampling efficiency of geometric correction of remote sensing image is very important.