Grassland degradation is an ecological problem facing the world. Investigating the species composition and species distribution of grassland is extremely important for judging the degradation process of grassland. At present, satellite remote sensing technology is difficult to meet the requirements of grassland species level classification due to the limitation of spatial resolution. Unmanned aerial vehicle (UAV) hyperspectral remote sensing technology provides images of centimeter level spatial resolution and nanoscale spectral resolution required for grassland species classification. Based on the UAV hyperspectral imaging remote sensing system, the hyperspectral image data of low and mixed growth desert grassland degradation indicator species were collected in the 400~1000nm spectral range. Flight experiments were carried out at the flowering, fruiting and yellow blight periods of the degraded indicator species. The flying height was 30m and the ground resolution of the hyperspectral image was about 2.3cm.Based on the combination of feature bands extraction and deep learning convolutional neural network (CNN), a method for classification of desert grassland species was proposed. The recommended phenological phase of species classification of desert grassland in central Inner Mongolia, China, was given in combination with plant phenology. The overall classification accuracy and Kappa coefficient reached 94% and 0.91, respectively. The results showed that the UAV hyperspectral imaging remote sensing technology and deep CNN can better classify the indicator species of desert grassland degradation. Compared with the support vector machine based on radial basis kernel function and the deep CNN based on principal component analysis, the deep CNN classification based on feature bands selection had the best effect and the highest classification accuracy. The method of CNN and the low-altitude remote sensing of UAV equipped with hyperspectral imager provided a new way to classify grassland species. The research result provided characteristic parameters for the judgment of grassland degradation succession process, and quantitative indicators for grassland ecological restoration management.