A Scene Classification Method for High Resolution Remote Sensing Images

Abstract

The invention relates to a method for classifying high-resolution remote sensing image scenes. According to the method, a local binary pattern (LBP) column diagram of a visual word distribution diagram is used for performing feature expression. The method includes the steps that a scene sub-image set is generated for set high-resolution remote sensing images, and a scene training sample set is established; dense grid sampling is performed on sub-images in the training sample set, local features are extracted, and K-means clustering is performed to construct a visual dictionary; visual word mapping is performed on all the sub-images in the scene training sample set to generate the visual word distribution diagram, and LBP conversion is performed on the visual word distribution diagram to obtain the LBP column diagram of the visual word distribution diagram; an SVM algorithm is used for generating a high-resolution remote sensing image scene classification model, and scene classification is performed on the high-resolution remote sensing images to be classified to obtain a scene classification diagram. The information utilization degree of the visual word distribution diagram is increased, and meanwhile the scene classification result diagram of the high-resolution remote sensing images to be classified can be obtained.

Description

technical field [0001] The invention relates to the technical field of high-resolution remote sensing image scene classification, in particular to a high-resolution remote sensing image scene classification method using a local binary pattern histogram of a visual word distribution map for feature expression. Background technique [0002] In recent years, with the further improvement of imaging resolution, more and more ultra-high-resolution remote sensing satellites are available, and the spatial information of the obtained high-resolution remote sensing images is more abundant, and the spatial relationship features become more specific. and detailed, the geometric structure and texture information of ground object targets are more obvious, the size of artificial targets is close to the size of high-resolution remote sensing image resolution, complex multi-coverage land use scenes under the influence of some human factors (such as airports, docks, etc.) , parking lot, etc.)...

AI Technical Summary

Problems solved by technology

However, the current research on scene classification of high-resolution remote sensing images is only for small image sets, and lacks the technology for scene classification of the entire remote sensing image; most of the visual word bag features used are based on visual word distribution maps. Overall histogram statistics or histogram statistics based on sub-regions are insufficient for visual word distribution maps

Images