GF-2 PMS2与ZY-3 MUX多光谱传感器数据的交互对比
|
摘要
近年来随着我国新型国产高分影像的相继问世以及相关应用的逐步展开,不少研究涉及了不同国产高分影像多光谱数据之间的交互对比,但两种国产分辨率最高的GF-2PMS2与ZY-3MUX传感器多光谱数据之间的对比仍未见报道。为了使这两种国产主力高分辨率传感器的多光谱数据能够在实际应用中相互补充使用,发挥更大的作用,基于它们的3对同日过空影像,采用两种方法对其进行交互对比。第一种方法是对整个试验区采用逐像元光谱比较法进行对比,第二种是采用样区光谱均值比较法进行对比,即在试验影像上选择一系列的样区,然后以各样区的均值进行对比。通过对两种传感器同步影像对的表观反射率进行回归分析,获得各对应波段的回归散点图,查明它们之间的定量关系,并据此提出相互转换的关系方程。研究结果表明,两种对比方法得出的结果相一致,但使用样区光谱均值比较法进行交互对比的结果的准确性更高。GF-2PMS2与ZY-3 MUX各对应波段具有很强的相关性,其线性回归方程的决定系数(R2)都大于0.9,但其值在蓝绿波段较高,在红光和近红外波段有所下降,表明两种传感器的表观反射率在蓝绿波段的一致性好于红光和近红外波段。总体上看,GF-2PMS2的信号强于ZY-3MUX,二者的信号差异在蓝、绿光波段较大,在红光和近红外波段较小,但却明显受到地物类型的影响。对于以裸土为主的影像,两种传感器之间的差异随着波长的增大而逐渐减小,而对于以植被为主的影像,二者之间的差异却随着波长的增大而逐渐增大。将纯植被与纯裸土的样区单独提取出来做进一步分析,结果表明,两种传感器的信号差异程度在红光波段主要受裸土影响,而在近红外波段则主要是受植被影响,且植被长势越旺盛,两种传感器的表观反射率差异越大。通过研究获得了两种传感器多光谱波段数据之间的相互转换方程,并对其进行验证,结果表明:经过转换后的GF-2PMS2数据与ZY-3MUX数据之间的差异大大减小,各波段均方根误差的均值降幅可达64.79%,平均相对偏差率也有明显的降低。这表明,所查明的两种传感器的定量关系是有效的,其对应波段的转换方程可以用于两种传感器数据的相互转换,经转换后的数据更有利于这两种传感器数据的协同使用。分析两种传感器数据的差异原因表明:二者数据的差异主要是由于它们的光谱响应函数的差异和空间分辨率的差异引起的。ZY-3 MUX的光谱响应函数曲线相对平缓,没有明显的起伏波动,而GF-2PMS2则较不稳定,在四个波段呈现出程度不同的起伏变化,从而影响了二者表观反射率信号的一致性;而GF-2PMS2具有的4m空间分辨率明显高于ZY-3MUX的6m空间分辨率,因此更容易捕捉到细小地物的光谱信息,这也使得二者信号出现不一致。
In recent years,China has launched a variety of Earth observation satellites with many newly-developed sensors onboard.Meanwhile,researches on the cross-comparison of these China-made new sensors are in progress.Nevertheless,no study has been published with respect to the comparison between Gaofen-2(GF-2)PMS2and Ziyuan-3(ZY-3)MUX sensor data up to now.The quantitative relationship between the two sensor data is unclear,and it is uncertain whether the two sensor data can be used for the same project directly.To meet this special requirement,this study carried out a cross-comparison between the GF-2PMS2 and ZY-3MUX sensor data based on three synchronous image pairs of the two sensors.The cross-comparison was performed using two methods.The first one is making use of image statistics based on large areas in common between the image pairs.A pixel-by-pixel comparison method was used to investigate quantitative relationship between GF-2PMS2 and ZY-3MUX sensor data based on the whole test area.The other method is a comparison based on the region of interest(ROI)in common to avoid the problem due to the difference in spatial resolution between the GF-2PMS2(4m)and ZY-3MUX(6m).The ROIs had appropriate size and were selected from homogeneous areas that excluded complicated terrain conditions.A linear regression model was adopted for the Top of Atmosphere(TOA)reflectance-based comparison between the ROIs of the GF-2PMS2 and ZY-3MUX images.Through the two methods,we obtained the quantitative relationship models between GF-2PMS2 and ZY-3MUX sensor data.This comparison study found that the results obtained by two methods,i.e.,pixel-by-pixel comparison and ROI-based comparison,are almost consistent.However,the ROI-based comparison achieves a higher accuracy because the spectral information of the corresponding pixels may be offset when using the pixel-by-pixel comparison due largely to the mis-registration of image pixels.This will lower the accuracy of the pixel-by-pixel comparison method.The results showed that the TOA reflectance of GF-2PMS2 and ZY-3MUX sensors has a high degree of agreement,with R2 values greater than 0.9for all the four bands.However,the higher R2 values in blue and green bands indicated that the TOA reflectance between the two sensors in both bands has a better agreement than that of red and near-infrared bands.Scatter plots showed that almost all data points lie under the one-to-one line in the spectral feature space with GF-2data in x-axis and ZY-3data in y-axis.This suggested that the GF-2PMS2 sensor data generally have higher TOA reflectance than ZY-3 MUX,especially in blue and green bands.It should be noted that the difference of TOA reflectance between the two sensor data can be affected by land cover types in red and near infrared bands.In the image pairs dominated by bare soil,the difference between the TOA reflectance of two sensors decreases with increasing wavelength,while for vegetation-dominated image pairs,the difference increases with increasing wavelength.In order to further examine the differences caused by the land cover types,more ROIs of pure vegetation and pure bare soil were extracted separately.The results showed that the signal difference between the two sensors is mainly affected by bare soil in the red band and by vegetation in the near infrared band.The more vigorous the vegetation grows,the greater the difference between the two sensors is.The band-by-band comparison has yielded the conversion equations for each corresponding bands of the two sensors,which were applied to convert the TOA reflectance between each corresponding bands of the two sensors.The validation of the conversion showed that the obtained conversion equations have high accuracy.It can be observed that the GF-2PMS2-simulated ZY-3MUX data are almost identical with the actual ZY-3 MUX data with R2 values close to 1and RMSE less than 0.01.The conversion has resulted in a significance reduction in RMSE by up to 64.79%,as well as a significant decrease in ME.This study showed that such a conversion can significantly improve the agreement between the two sensors data.The converted data are more conducive to the synergy between the GF-2PMS2 and ZY-3 MUX sensor data.The analysis showed that the differences in TOA reflectance between the two sensor data result probably from the differences in their spectral response function and spatial resolution.We found that the spectral response curve of ZY-3 MUX is smoother with no obvious fluctuations than that of GF-2PMS2,which is fluctuant in all of four bands.Such a difference in the spectral response functions may have led to the difference in TOA reflectance between the two sensors.In addition,the spatial resolution of GF-2PMS2is4 m,which is higher than ZY-3MUX's 6m.A higher spatial resolution will help GF-2PMS2 sensor to detect subtle spectral information of small ground objects and thus cause the difference in TOA reflectance between the two sensors.
In recent years,China has launched a variety of Earth observation satellites with many newly-developed sensors onboard.Meanwhile,researches on the cross-comparison of these China-made new sensors are in progress.Nevertheless,no study has been published with respect to the comparison between Gaofen-2(GF-2)PMS2and Ziyuan-3(ZY-3)MUX sensor data up to now.The quantitative relationship between the two sensor data is unclear,and it is uncertain whether the two sensor data can be used for the same project directly.To meet this special requirement,this study carried out a cross-comparison between the GF-2PMS2 and ZY-3MUX sensor data based on three synchronous image pairs of the two sensors.The cross-comparison was performed using two methods.The first one is making use of image statistics based on large areas in common between the image pairs.A pixel-by-pixel comparison method was used to investigate quantitative relationship between GF-2PMS2 and ZY-3MUX sensor data based on the whole test area.The other method is a comparison based on the region of interest(ROI)in common to avoid the problem due to the difference in spatial resolution between the GF-2PMS2(4m)and ZY-3MUX(6m).The ROIs had appropriate size and were selected from homogeneous areas that excluded complicated terrain conditions.A linear regression model was adopted for the Top of Atmosphere(TOA)reflectance-based comparison between the ROIs of the GF-2PMS2 and ZY-3MUX images.Through the two methods,we obtained the quantitative relationship models between GF-2PMS2 and ZY-3MUX sensor data.This comparison study found that the results obtained by two methods,i.e.,pixel-by-pixel comparison and ROI-based comparison,are almost consistent.However,the ROI-based comparison achieves a higher accuracy because the spectral information of the corresponding pixels may be offset when using the pixel-by-pixel comparison due largely to the mis-registration of image pixels.This will lower the accuracy of the pixel-by-pixel comparison method.The results showed that the TOA reflectance of GF-2PMS2 and ZY-3MUX sensors has a high degree of agreement,with R2 values greater than 0.9for all the four bands.However,the higher R2 values in blue and green bands indicated that the TOA reflectance between the two sensors in both bands has a better agreement than that of red and near-infrared bands.Scatter plots showed that almost all data points lie under the one-to-one line in the spectral feature space with GF-2data in x-axis and ZY-3data in y-axis.This suggested that the GF-2PMS2 sensor data generally have higher TOA reflectance than ZY-3 MUX,especially in blue and green bands.It should be noted that the difference of TOA reflectance between the two sensor data can be affected by land cover types in red and near infrared bands.In the image pairs dominated by bare soil,the difference between the TOA reflectance of two sensors decreases with increasing wavelength,while for vegetation-dominated image pairs,the difference increases with increasing wavelength.In order to further examine the differences caused by the land cover types,more ROIs of pure vegetation and pure bare soil were extracted separately.The results showed that the signal difference between the two sensors is mainly affected by bare soil in the red band and by vegetation in the near infrared band.The more vigorous the vegetation grows,the greater the difference between the two sensors is.The band-by-band comparison has yielded the conversion equations for each corresponding bands of the two sensors,which were applied to convert the TOA reflectance between each corresponding bands of the two sensors.The validation of the conversion showed that the obtained conversion equations have high accuracy.It can be observed that the GF-2PMS2-simulated ZY-3MUX data are almost identical with the actual ZY-3 MUX data with R2 values close to 1and RMSE less than 0.01.The conversion has resulted in a significance reduction in RMSE by up to 64.79%,as well as a significant decrease in ME.This study showed that such a conversion can significantly improve the agreement between the two sensors data.The converted data are more conducive to the synergy between the GF-2PMS2 and ZY-3 MUX sensor data.The analysis showed that the differences in TOA reflectance between the two sensor data result probably from the differences in their spectral response function and spatial resolution.We found that the spectral response curve of ZY-3 MUX is smoother with no obvious fluctuations than that of GF-2PMS2,which is fluctuant in all of four bands.Such a difference in the spectral response functions may have led to the difference in TOA reflectance between the two sensors.In addition,the spatial resolution of GF-2PMS2is4 m,which is higher than ZY-3MUX's 6m.A higher spatial resolution will help GF-2PMS2 sensor to detect subtle spectral information of small ground objects and thus cause the difference in TOA reflectance between the two sensors.
引文
[1] XIN Rui,LU Zhong-jun,LIU Yang,et al(辛蕊,陆忠军,刘洋,等).Transactions of the Chinese Society of Agricultural Engineering(农业工程学报),2015,31(16):196.
[2] YANG Lei,FU Qiao-yan,PAN Zhi-qiang,et al(杨磊,傅俏燕,潘志强,等).Infrared and Laser Engineering(红外与激光工程),2015,44(8):2456.
[3] Chen Yifu,Xie Zhong,Qiu Zhengge,et al.IEEE Transactions on Geoscience and Remote Sensing,2015,53(8):4616.
[4] Holden C E,Woodcock C E.Remote Sensing of Environment,2016,185:16.
[5] HAN Qi-jin,PAN Zhi-qiang,ZHANG Xue-wen,et al(韩启金,潘志强,张学文,等).Acta Optica Sinica(光学学报),2014,34(6):0628003.
[6] JIA Yu-qiu,LI Bing,CHENG Yong-zheng,et al(贾玉秋,李冰,程永政,等).Transactions of the Chinese Society of Agricultural Engineering(农业工程学报),2015,31(9):173.
[7] Feng L,Li J,Gong W,et al.Remote Sensing of Environment,2016,174:56.
[8] SONG Jun-wei,ZHANG You-jing,LI Xin-chuan,et al(宋军伟,张友静,李鑫川,等).Progress in Geography(地理科学进展),2016,35(2):255.
[9] XU Han-qiu,LIU Zhi-cai,GUO Yan-bin(徐涵秋,刘智才,郭燕滨).Transactions of the Chinese Society of Agricultural Engineering(农业工程学报),2016,32(8):148.
[10] XU Han-qiu(徐涵秋).Geomatics and Information Science of Wuhan University(武汉大学学报·信息科学版),2007,32(1):62.
[11] Chander G,Markham B L,Helder D L.Remote Sensing of Environment,2009,113(5):893.
[12] XU Han-qiu,ZHANG Tie-jun,HUANG Shao-lin(徐涵秋,张铁军,黄绍霖).Geographical Research(地理研究),2013,32(7):1336.
[13] Li H,Chen Z X,Jiang Z W,et al.Journal of Integrative Agriculture,2017,16(2):266.
[14] Wang L,Yang R,Tian Q,et al.Remote Sensing,2015,7(2):2089.
[2] YANG Lei,FU Qiao-yan,PAN Zhi-qiang,et al(杨磊,傅俏燕,潘志强,等).Infrared and Laser Engineering(红外与激光工程),2015,44(8):2456.
[3] Chen Yifu,Xie Zhong,Qiu Zhengge,et al.IEEE Transactions on Geoscience and Remote Sensing,2015,53(8):4616.
[4] Holden C E,Woodcock C E.Remote Sensing of Environment,2016,185:16.
[5] HAN Qi-jin,PAN Zhi-qiang,ZHANG Xue-wen,et al(韩启金,潘志强,张学文,等).Acta Optica Sinica(光学学报),2014,34(6):0628003.
[6] JIA Yu-qiu,LI Bing,CHENG Yong-zheng,et al(贾玉秋,李冰,程永政,等).Transactions of the Chinese Society of Agricultural Engineering(农业工程学报),2015,31(9):173.
[7] Feng L,Li J,Gong W,et al.Remote Sensing of Environment,2016,174:56.
[8] SONG Jun-wei,ZHANG You-jing,LI Xin-chuan,et al(宋军伟,张友静,李鑫川,等).Progress in Geography(地理科学进展),2016,35(2):255.
[9] XU Han-qiu,LIU Zhi-cai,GUO Yan-bin(徐涵秋,刘智才,郭燕滨).Transactions of the Chinese Society of Agricultural Engineering(农业工程学报),2016,32(8):148.
[10] XU Han-qiu(徐涵秋).Geomatics and Information Science of Wuhan University(武汉大学学报·信息科学版),2007,32(1):62.
[11] Chander G,Markham B L,Helder D L.Remote Sensing of Environment,2009,113(5):893.
[12] XU Han-qiu,ZHANG Tie-jun,HUANG Shao-lin(徐涵秋,张铁军,黄绍霖).Geographical Research(地理研究),2013,32(7):1336.
[13] Li H,Chen Z X,Jiang Z W,et al.Journal of Integrative Agriculture,2017,16(2):266.
[14] Wang L,Yang R,Tian Q,et al.Remote Sensing,2015,7(2):2089.