Research on ground information identification method for far-field remote sensing images based on transfer learning

In recent years, with the vigorous development and widespread application of artificial intelligence (AI) technology, AI has empowered remote sensing technology, significantly shortening the cycle of remote sensing image interpretation, improving accuracy, and promoting the transformation of remote sensing technology applications. However, the interpretability and generalization transfer ability of some existing models are average, and the results have obvious limitations. This article focuses on the problem of ground information interpretation based on satellite remote sensing images, and conducts research on hyperspectral image classification methods based on EfficientNet deep learning. Furthermore, the EfficientNet transfer learning model is optimized based on an improved GJO. While extracting shallow features such as edges and textures, the problem of high intra class differences and low inter class differences in hyperspectral remote sensing images is effectively addressed, which can effectively improve the efficiency of satellite remote sensing image land recognition. Finally, the comparative classification experiment of deep transfer learning also proves the progressiveness of the method proposed in this paper.