Spatial–Spectral Fusion of HY-1C COCTS/CZI Data for Coastal Water Remote Sensing Using Deep Belief Network

Abstract

The remote sensing monitoring of coastal waters with dramatic changes requires images with high spatial and temporal resolutions and adequate spectral bands. However, a single sensor is limited to meet these requirements. Image fusion is, therefore, widely adopted. In this article, a deep belief network (DBN) is developed to fuse images from the Chinese ocean color and temperature scanner (1000 m, eight bands) and coastal zone imager (50 m, four bands) onboard HaiYang-1C satellite to generate 50-m, eight-band, and three-day observations for coastal waters. The DBN is compared with the existing prevailing Gram–Schmidt transformation (GS) and inversion-based fusion (IBF) algorithms over the Bohai Sea at the top-of-atmosphere reflectance and product [e.g., chlorophyll-a (chl-a)] levels. Results indicate that for the spatial aspect, DBN can avoid the block effect and maintain details. The average structural similarity index of DBN is approximately 22.08% and 3.30% better than that of GS and IBF, respectively; for the spectral aspect, the mean relative errors for eight bands of DBN range from 3.15% to 21.54%. The errors are less than 50% and 80% of those of GS, while less than 80% and 110% of those of IBF, at bands 1–6 and bands 7 and 8, respectively; for chl-a retrieval, DBN yields better results with the coefficient of determination R 2 of 0.78 and root-mean-square error (RMSE) of 0.10 mg/m3 compared with those of IBF ( R 2 = 0.59 and RMSE = 0.16 mg/m3). DBN outperforms GS and IBF at reflectance and product levels, displaying great potential for the remote sensing monitoring of coastal waters.