Abstract: Aiming at the problem that the current spatial database clustering method did not consider the inverse result of the distance feature after dimensionality reduction, which led to the distortion of spatial data components, the clustering accuracy was low and the timeconsuming was long, the author proposed a reverse nearest neighbor clustering method of spatial database. Firstly, the feature decomposition of the kernel matrix was realized by selecting the training sample set, and the distance 
feature correction value was obtained to remove the influence of initial value. Secondly, according to dimensionality reduction of kernel principal component analysis (KPCA) combined with the inverse result of distance feature after dimensionality reduction, the inverse nearest neighbor clustering method was combined with the extended partial distortion search method to realize spatial data clustering. Finally, the data set was calculated by using the selected clustering center, the distortion between the first dimension component of the data set and the first dimension component of the clustering center was calculated, the reverse nearest neighbor was obtained until all spatial data found the category, and the reverse nearest neighbor clustering of spatial database was completed. The experimental results show that the method improves the clustering accuracy of spatial data and reduces the time used for spatial data clustering.

Key words: spatial database, spatial distance, data correction, dimensionality reduction, reverse nearest neighbor, clustering method