Remote sensing image scene classification has high application value in agriculture, land planning and other fields, and has become a hot topic in current research.However, in practical applications, remote sensing data can not be obtained in one time. How to classify remote sensing images presented in the form of data streams becomes a common problem in the field of remote sensing. The existing scene classification dataset lacks a unified standard for data streams division,which largely limits the development of scene classification algorithms for data streams. Firstly, this paper gives a set of remote sensing image scene classification database of data streams construction and evaluation criteria, and proposes a large-scale remote sensing image scene classification database CLRS, which is used to evaluate and improve the remote sensing image scene classification algorithm of data streams. In addition, this paper also introduces continuous learning into the field of remote sensing.In the remote sensing image classification scenarios of three data streams, the current mainstream continuous learning method is tested and analyzed, which can be used as the baseline results for future development the scene classification algorithm of data streams.