Cloud detection is a crucial process of satellite remote sensing because of its significant impacts on the monitoring and quantitative retrieval of various atmospheric and surface parameters, and its accuracy also affects the retrieval of cloud microphysical properties. In this study, an improved cloud detection algorithm based on spectral information is established for the directional polarimetric camera (DPC) onboard Gaofen-5 satellite. Based on thedifferences between cloud pixels and clear pixels in the visible spectrum, the absorption characteristics of oxygen A-band, the polarized reflectance of blue band, and the polarized rainbow feature, the algorithm establishes a comprehensive cloud detection method over land and ocean area, and further proposes a multi-angle fusion strategy for the final determination of cloud detection. The recognition of high thin clouds and low thin clouds is improved by adding apparent pressure detection and polarization rainbow detection over land, respectively, and in other hand, the combination of apparent pressure and degree of linear polarization is used to improve the detection of cloud pixels in the sun-glint area over ocean. Examples of global cloud detection results of DPC have shown the effectiveness of this method, and the results of typical areas have also shown good consistency with MODIS cloud products. This research can be expected to provide a methodological basis for the cloud detection on the multi-angle and polarimetric sensor onboard the recently launched Gaofen-5(02) satellite.