The appearance of hyperspectral remote sensing image further improves the accuracy of remote sensing image classification, but the data of hyperspectral remote sensing image is large, and the processing hyperspectral remote sensing image has high complexity and low efficiency. A remote sensing image classification algorithm based on improved bilinear recurrent neural network (BLRNN) model is proposed, this paper gives the definition of bilinear recurrent neural Network and the description of network structure, and optimizes and improves the bilinear recurrent neural network from two aspects of network structure and pruning process, and uses the genetic algorithm global search to trim. Compared with the original feature, PCA and BPNN algorithm, the results show that the BLRNN algorithm has been greatly improved in classification accuracy and classification time, and the image processing efficiency is improved.