Abstract: A spatial semantic model based method is proposed to solve the issue of automatically detecting geo-objects in high resolution remote sensing images. This method obtains firstly image segments through over-segmentation algorithm, and calculates the multiple features by using topic models, in order to improve the description accuracy of segments attribution. Then, this method investigates and models the spatial relationship between geo-objects in whole images, and a semantic parsing tree of the scene category is extracted, which could be used to detect and locate the geo-objects. The experimental results on the dataset demonstrate the robustness and accuracy of this method.