轻小型无人机高光谱影像拼接研究
|
摘要
轻小型、低成本无人机(unmanned aerial vehicle, UAV)机载光谱成像仪的快速发展为水质监测、精准农业提供了新的手段。ZK-VNIR-FPG480机载高光谱成像仪是国产仪器,拥有自主产权,影像共有270个波段,光谱范围为400～1 000 nm,光谱分辨率为3 nm,空间分辨率为0.9 m@1 km,成像方式为运动推扫成像,该成像仪的特点是影像之间不存在航向重叠,只存在旁向重叠。它在提供高光谱、高空间分辨率影像的同时也存在着一系列问题:①无人机的狭窄视场限制了每条航带的地面覆盖范围,需要进行航带拼接;②其自带的POS系统定位精度低;③为提高作业效率,航带间的重叠率较低,一般设置在30%左右,为影像拼接增加了困难;④因飞行时受风力、光照以及仪器自身等影响使得每条航带间存在亮度差异,拼接时会出现拼接缝现象。针对上述问题提出一种基于曲面样条函数和相位相关的无人机高光谱影像拼接方法,旨在将无人机拍摄的单条高光谱航带拼接成一幅完整的带有地理坐标的全景图,并实现影像几何和光谱上的匹配。该方法包括以下几个步骤:首先,以正射影像为基准采用曲面样条函数法对高光谱航带进行地理配准,赋予每条航带真实的地理坐标;然后采用局部方差法计算各波段信噪比,取分值最高的波段作为最优波段;再利用该最优波段采用基于2幂子图像的相位相关算法来纠正航带间已经存在的地理空间映射关系,消除航带间存在的错位;最后选用加权平均融合法对相邻航带进行融合,消除航带拼接时因光照、仪器自身等影响所产生的拼接线问题,最终得到带有绝对地理坐标的高光谱全景图。实验使用ZK-VNIR-FPG480机载高光谱成像仪获取大理某地区的高光谱数据进行拼接,结果表明,该拼接方法得到的全景图拼接处没有错位现象,几何位置准确。选取4种典型地物拼接前后的光谱曲线,其曲线走向基本一致,计算拼接影像与拼接前左右影像的光谱角余弦均值为0.965 2,光谱相关系数均值为0.863 2,光谱信息散度均值为0.424 0,欧式距离均值为0.494 1,四种光谱曲线相似性测度指标客观上显示了曲线的高度相似性,表明拼接前后同名点的光谱匹配度高,适用于无人机高光谱数据的拼接。该方法不仅提高了拼接影像的地理坐标精度,还在消除拼接缝的基础上最大限度的保证了光谱的保真性,并通过引入2幂子图像解决了影像在重叠度低的情况下配准算法失效的问题。但拼接前相邻航带同名点间存在光谱差异,且高光谱数据量大,拼接耗时多,如何利用重叠区域的像素修正系统误差,统一拼接图像的度量空间以提升光谱精度和稳定性并提高拼接速度仍是今后需要解决的问题。
The rapid development of small and low-cost unmanned aerial imaging spectrometer has provided new means for water quality monitoring and precision agriculture. The ZK-VNIR-FPG480 airborne hyperspectral imager is a domestic development instrument with independent property right. The image has a total of 270 bands, the spectral range is 400～1 000 nm, the spectral resolution is 3 nm, and the spatial resolution is 0.9 m@1 km. The imaging method is motion push broom imaging, which is characterized by no overlap between the images and only overlapping. While providing high spectral and high spatial resolution images, it also has a series of problems: ①the narrow field of unmanned aerial vehicle(UAV) restricts the coverage of the ground surface of each airstrip, and requires the splicing of flights; ②the positioning accuracy of its POS system is low; ③In order to improve operational efficiency, the overlap ratio between navigation bands is relatively low, which is generally set at about 30%, making it difficult to image splicing; ④Due to the influence of wind, light, and the instrument itself during flight, there is a difference in brightness between each band, and stitching occurs when stitching occurs. This paper proposes a method for splicing UAV hyperspectral images based on surface spline function and phase correlation, aiming at solving the above problems. The aim is to splice a single hyperspectral band taken by UAV into a complete panorama with geographic coordinates, and to achieve image geometry and spectral matching. The method includes the following steps: First, the hyperspectral flight is georeferenced using the surface spline function method with the orthophoto image as the reference, and the real geographic coordinates of each flight are assigned; second, the local variance method is used to calculate the signal-to-noise ratio of each band, and the highest value band is taken as the optimal band; and then the phase correlation algorithm based on the 2 power image is used to correct the existing geographic spatial mapping relations between flights and eliminate the dislocation of the flight. Finally, the weighted average fusion method is used to fuse the adjacent flights and eliminate the problem of the mosaic line caused by the illumination and the instrument itself. Through the above steps, we can get a hyperspectral panorama with absolute geographic coordinates. The experiment uses ZK-VNIR-FPG480 airborne hyperspectral imager to get the hyperspectral data of a region of Dali to splice. The results show that the splicing method has no dislocations in the panorama stitching, and the geometric position is accurate. The curve directions of the 4 typical objects before and after splicing are basically the same. The average value of the spectral cosine of the left and right images before and after the stitching image was calculated to be 0.965 2, the average value of the spectral correlation coefficient was 0.863 2, the average value of the spectral information divergence was 0.424 0, and the average value of the Euclidean distance was 0.494 1. The four kinds of spectral curve similarity measure indicators objectively showed the high similarity of the curves, indicating that the spectral matching degree of the same name point before and after splicing is high, which is suitable for the splicing of UAV hyperspectral data. The method not only improves the accuracy of the geographical coordinates of the spliced image, but also ensures the maximum spectrum fidelity on the basis of the elimination of the joint joint. The 2 power image is introduced to solve the problem of the registration algorithm failure under the low overlap of the image. However, there is a spectral difference between the same name points in adjacent bands before splicing, and the amount of hyperspectral data is large and the splicing takes more time. It is still a problem to figure out how to use the pixels of the overlapped region to correct the system error, to unify the measurement space of the image, to improve the spectral accuracy and stability and to improve the stitching speed.
The rapid development of small and low-cost unmanned aerial imaging spectrometer has provided new means for water quality monitoring and precision agriculture. The ZK-VNIR-FPG480 airborne hyperspectral imager is a domestic development instrument with independent property right. The image has a total of 270 bands, the spectral range is 400～1 000 nm, the spectral resolution is 3 nm, and the spatial resolution is 0.9 m@1 km. The imaging method is motion push broom imaging, which is characterized by no overlap between the images and only overlapping. While providing high spectral and high spatial resolution images, it also has a series of problems: ①the narrow field of unmanned aerial vehicle(UAV) restricts the coverage of the ground surface of each airstrip, and requires the splicing of flights; ②the positioning accuracy of its POS system is low; ③In order to improve operational efficiency, the overlap ratio between navigation bands is relatively low, which is generally set at about 30%, making it difficult to image splicing; ④Due to the influence of wind, light, and the instrument itself during flight, there is a difference in brightness between each band, and stitching occurs when stitching occurs. This paper proposes a method for splicing UAV hyperspectral images based on surface spline function and phase correlation, aiming at solving the above problems. The aim is to splice a single hyperspectral band taken by UAV into a complete panorama with geographic coordinates, and to achieve image geometry and spectral matching. The method includes the following steps: First, the hyperspectral flight is georeferenced using the surface spline function method with the orthophoto image as the reference, and the real geographic coordinates of each flight are assigned; second, the local variance method is used to calculate the signal-to-noise ratio of each band, and the highest value band is taken as the optimal band; and then the phase correlation algorithm based on the 2 power image is used to correct the existing geographic spatial mapping relations between flights and eliminate the dislocation of the flight. Finally, the weighted average fusion method is used to fuse the adjacent flights and eliminate the problem of the mosaic line caused by the illumination and the instrument itself. Through the above steps, we can get a hyperspectral panorama with absolute geographic coordinates. The experiment uses ZK-VNIR-FPG480 airborne hyperspectral imager to get the hyperspectral data of a region of Dali to splice. The results show that the splicing method has no dislocations in the panorama stitching, and the geometric position is accurate. The curve directions of the 4 typical objects before and after splicing are basically the same. The average value of the spectral cosine of the left and right images before and after the stitching image was calculated to be 0.965 2, the average value of the spectral correlation coefficient was 0.863 2, the average value of the spectral information divergence was 0.424 0, and the average value of the Euclidean distance was 0.494 1. The four kinds of spectral curve similarity measure indicators objectively showed the high similarity of the curves, indicating that the spectral matching degree of the same name point before and after splicing is high, which is suitable for the splicing of UAV hyperspectral data. The method not only improves the accuracy of the geographical coordinates of the spliced image, but also ensures the maximum spectrum fidelity on the basis of the elimination of the joint joint. The 2 power image is introduced to solve the problem of the registration algorithm failure under the low overlap of the image. However, there is a spectral difference between the same name points in adjacent bands before splicing, and the amount of hyperspectral data is large and the splicing takes more time. It is still a problem to figure out how to use the pixels of the overlapped region to correct the system error, to unify the measurement space of the image, to improve the spectral accuracy and stability and to improve the stitching speed.
引文
[1] Wang C,Guo Z,Wang S,et al.Journal of Spectroscopy,2015,2015:ID969185.
[2] Asadzadeh S,Roberto D S F C.International Journal of Applied Earth Observations & Geoinformation,2016,47:69.
[3] XU Zhi-fang,WANG Shuang-ting,WANG Chun-yang,et al(许志方,王双亭,王春阳,等).Remote Sensing Information(遥感信息),2017,32(1):95.
[4] Kim J I,Kim T,Shin D,et al.International Journal of Remote Sensing,2017,38(8-10):2557.
[5] CHENG Yuan-hang(程远航).Research on Aerial Remote Sensing Image Mosaic Technology for UAV(无人机航空遥感图像拼接技术研究).Beijing:Tsinghua University Press(北京:清华大学出版社),2016.7.
[6] Ren X,Sun M,Zhang X,et al.Remote Sensing,2017,9(9):962.
[7] GONG Sheng-sheng(龚声胜).Theoretical Research in Urban Construction(城市建设理论研究:电子版),2015,5(15):5284.
[8] ZHANG Lu,HUANG Wu-meng,ZHANG Yue,et al(张璐,黄吴蒙,张悦,等).Geospatial Information(地理空间信息),2013,11(3):73.
[9] ZHOU Mei-li,BAI Zong-wen,YAN Xiao-jin(周美丽,白宗文,延小进).Foreign Electronic Measurement Technology(国外电子测量技术),2015,34(5):31.
[10] Pant M,Ray K,Sharma T K,et al.Soft Computing:Theories and Applications,Proc.SoCTA 2016,Vol.2 (Advances in Intelligent Systems and Computing 584).Springer,2017.
[11] ZHANG Jun-zhe,ZHU Wen-quan,ZHENG Zhou-tao,et al(张浚哲,朱文泉,郑周涛,等).Science of Surveying and Mapping(测绘科学),2013,38(6):33.
[2] Asadzadeh S,Roberto D S F C.International Journal of Applied Earth Observations & Geoinformation,2016,47:69.
[3] XU Zhi-fang,WANG Shuang-ting,WANG Chun-yang,et al(许志方,王双亭,王春阳,等).Remote Sensing Information(遥感信息),2017,32(1):95.
[4] Kim J I,Kim T,Shin D,et al.International Journal of Remote Sensing,2017,38(8-10):2557.
[5] CHENG Yuan-hang(程远航).Research on Aerial Remote Sensing Image Mosaic Technology for UAV(无人机航空遥感图像拼接技术研究).Beijing:Tsinghua University Press(北京:清华大学出版社),2016.7.
[6] Ren X,Sun M,Zhang X,et al.Remote Sensing,2017,9(9):962.
[7] GONG Sheng-sheng(龚声胜).Theoretical Research in Urban Construction(城市建设理论研究:电子版),2015,5(15):5284.
[8] ZHANG Lu,HUANG Wu-meng,ZHANG Yue,et al(张璐,黄吴蒙,张悦,等).Geospatial Information(地理空间信息),2013,11(3):73.
[9] ZHOU Mei-li,BAI Zong-wen,YAN Xiao-jin(周美丽,白宗文,延小进).Foreign Electronic Measurement Technology(国外电子测量技术),2015,34(5):31.
[10] Pant M,Ray K,Sharma T K,et al.Soft Computing:Theories and Applications,Proc.SoCTA 2016,Vol.2 (Advances in Intelligent Systems and Computing 584).Springer,2017.
[11] ZHANG Jun-zhe,ZHU Wen-quan,ZHENG Zhou-tao,et al(张浚哲,朱文泉,郑周涛,等).Science of Surveying and Mapping(测绘科学),2013,38(6):33.