Accurate classification of the land cover from the digital surface model (DSM) obtained from LiDAR sensors is a challenging topic thatwhich has been considered by researchers in recent years. In general, the classification accuracy of land cover leads to low accuracy using a single-band DSM image, hence, the needit seems necessary to develop efficient methods to extract suitable spatial information thatwhich improves classification accuracy seems necessaryaccuracy. In this regard, the use of spatial features based on morphological profiles (MPs) has been proven to increase the accuracy of classification significantly in recent years. Despite the efficiency of MPs in increasing the classification accuracy of the LiDAR-DSM, the proposed classification methods cannot effectively consider the local relationships between the features. In the present paper, we try to improve the classification accuracy by using local weighted kernel descriptors thatwhich are capable of considering high-order statistics. Experimental results on the well-known Houston University data set show that the proposed method increases the classification accuracy by about 13%, which is also statistically significant. In addition, the proposed method outperformed two other recent DSM classification methods by about 3%.