With the influence of climate change and human activities,the grasslands in Inner Mongolia are gradually desertifying.In order to solve the limitation of traditional ground survey and the problem of difficulty to classify small samples of hyperspectral data,in this paper,we use unmanned aerial vehicle (UAV) hyperspectral remote sensing technology to collect data on desert grassland species and propose a small sample classification method with a neighborhood aggregation algorithm combined with deep learning.Firstly,the band selection of hyperspectral data is performed by a genetic algorithm (GA) with the optimum index factor (OIF).Secondly,the neighborhood features of hyperspectral data are constructed,and the neighborhood aggregation algorithm is used to fuse the neighborhood features.Finally, the fused features are classified by a multilayer perceptron (MLP).The results show that the overall accuracy of the neighborhood aggregation algorithm can reach 93.41% with only 10 samples per class of features,and the Kappa coefficient is 0.912 0;compared with SVM and various deep learning models,the neighborhood aggregation algorithm has high computational efficiency,the smallest model size,and the highest classification accuracy.The proposed method meets the requirements of grassland species identification and provides a new method for the dynamic monitoring of grassland ecosystems.