Research On Multi-object Retrieval Of Remotely Sensed Images Based On Spatial Relationship

With the concept of the internet of things and smarter planet proposed, remote sensing images play an increasingly important role in people's daily life. However, the content of remote sensing images is complicated and various. As a result, quick obtaining of desired information from remote sensing image data becomes a recognized difficult problem. Content-based image retrieval technology– which realizes image retrieval by analyzing the image content (including color, texture, shape, objects and spatial relationships, context and semantics, etc.) - provides an excellent solution for remote sensing image retrieval.The development of space exploration and communication technology accelerates the growth of remote sensing image data of high resolution at amazing speed every day. Retrieval on object level as well as space relationship level and identification to complicated objects– remote sensing image of high resolution– are full of high importance.This paper studies the key technologies concerning the retrieval of remote sensing images at first. Those technologies contain the organizing and management methods of remote sensing image data, the extraction and expression of image feature, the measurement of characteristic similarity, the evaluation standard of retrieval performance, and so on. In succession, further research is performed on the classification and expression of spatial relations. A variety of representation models about spatial relations are introduced, such as the topological expression of point set, the projection interval expression, the expression of the direction relations and force histogram, etc. Next, this paper focuses on retrieval methods of the spatial-relations-based remote sensing images. These methods consist of the ones based on texture segmentation and projection interval relationship, based on force histogram scale and rotation invariant multi-object orientation relationship, based on integration of texture and spatial relationships and so on. Through experimental results and quantitative analysis on the retrieval performance, the feasibility of these retrieval methods depicted in this article are fully confirmed and consolidated. Finally, this paper explores the design of structure and function modules of remote sensing image retrieval prototype system. Research results suggest that the use of spatial relationship in remote sensing image data retrieval makes better retrieval performance. The remote sensing image retrieval, which integrates texture and spatial relationships, can effectively compensate for the insufficiency caused by using single spatial relationship, and accordingly obtains better retrieval effect. This research has theoretical and application value on extraction of remote sensing image information, sharing and broadening remote sensing image data.