Hyper spectral image classification has become one of the most important research directions in detection technology; furthermore, it has been widely used in military and civilian fields. However, the significant number of bands, data redundancy, and low utilization of spatial features render the classification of hyper spectral images challenging, and most of existing hyper spectral image classifications use visible light or short-wave infrared data. Hence, a K-means classification method based on spectral and spatial features is proposed in this paper. First, spatial features are extracted; next, the spectral features are combined with the spatial features and the dimensions are reduced. Finally, the K-means algorithm is introduced to obtain classification results that are better than those of normal K-means, and the algorithm is applied to long-wave infrared hyper spectral image classification.