Automatic Recognition of Landscape Linear Features from High-Resolution Satellite Images

The development of high-resolution sensors provided more methods for remote sensing mapping. High-resolution satellite imagery refers to those images of which spatial resolution is less than 10m. Information of linear features of landscape in high-resolution satellite images is very prolific. However, every part of a whole scene has different characters in spectrum space or gray space, because of immoderate specializations and influence of noise disturbances. This made it difficult to detect and recognize the linear features of landscape objects. The traditional single filtering method used to extract linear features of objects was confronted with some problems such as feature specializations. BRDF influences, geo-procedure scaling dependence etc. Method about extraction of structure information from high-resolution imagery is a key for application of this imagery. In this paper, we discussed a method based on image texture characters through designing filter cluster for the automatic detection and recognition. The filter cluster includes two filters, one is the high pass filter through which the borderline characteristics of the object in imagery can be detected, another is the statistical filter through which some noise made by high pass filter can be abated. In process of high pass filtering, how to select the size of filtering window is very important to precision. We had designed a proportion factor η to depict processing result. The test result indicated that the 7×7 filter window was the most appropriate in this research. After being transacted by the filter cluster, through selection of threshold on the test image which could control dividing pixels of background and objects, a vector layer that describing the linear features of object in landscape could be acquired through linear detection and tracking. A comparison of differences was finished between the results by traditional single filter and by multi-filter. The vector layer gained through multi-filtering has fewer noises than those only through single filtering. The result of test indicated that multi-filtering method could improve the precision of analysis, illustrate the possiblity of extracting automa- tically the vector information layers from high resolution satellite imagery, and provided technology basis for application of high-resolution satellite imagery.