“The "data+knowledge" dual drive theoretical framework achieves higher levels of intelligence through data-driven perceptual ability and knowledge driven cognitive ability, breaking through the bottleneck of generalization and interpretability in a single data-driven paradigm. The research progress in remote sensing image feature extraction is introduced, and the team systematically reviews traditional and deep learning methods, summarizes three types of knowledge guidance paradigms: visual knowledge, geographic spatial knowledge, and physical knowledge, laying the foundation for the construction of remote sensing intelligent interpretation system.”