渤海浑浊水体GOCI影像神经网络大气校正研究

针对GOCI遥感数据的官方处理软件(GDPS(GOCI data processing system))中的标准大气校正算法在渤海近岸浑浊水体区域处理中存在的问题,以MODIS/Aqua数据NIR-SWIR波段联合大气校正处理所得的水色遥感反射率产品结果和GOCI的星上反射率数据为基础,基于神经网络模型进行浑浊水体GOCI影像的大气校正方法研究结果表明,神经网络方法能显著减少标准产品中大气校正失效区域,特别是在443、490、680、555、745nm波段改进效果非常明显;但412、660、865nm波段的紧邻近岸的浑浊水体部分区域存在遥感反射率空间分布不合理,这可能与MODIS对应波段产品本身的大气校正精度不高有关.由于缺乏对应的实测数据,后续验证工作还需要进一步开展.

Atmospheric correction of GOCI imagery over turbid waters in Bohai Sea based on artificial neural network

The standard GOCI atmospheric correction algorithm was designed for open ocean water, embedded in GOCI data processing system（GDPS）. In allusion to the failure of standard algorithm over coastal turbid waters in Bohai Sea, neural network model-based atmospheric correction method was proposed for the processing of GOCI images for the turbid waters. The neural network was built by using the top of atmosphere reflectance of GOCI as input, and the remote sensing reflectance in visible and near-infrared bands of MODIS/Aqua from NIR-SWlR atmospheric correction algorithm as output. The results showed that the proposed method notably decreased failure regions of standard atmospheric correction algorithm products, especially in 443,490,680,555 and 745 nm bands, in spite of the less well results of the spatial distribution of remote sensing reflectance in 412, 660 and 865 nm bands. Due to lacking of corresponding in-situ data, subsequent work of verification will be carried out in the future.