Abstract:

A spatiallysensitive bags of feature （SS-BOF） is introduced which is used for terrain recognition of synthetic aperture radar (SAR) images. Firstly，a generalized nuclear fuzzy CMeans method is used to segment the SAR images, then the target shape of each SAR image is extracted. Secondly, its corners are extracted by using Harris corner way, then the target graph is triangulated by Delaunay triangulation. Thirdly, the cotangent weight is assigned to the triangulation, then the eigenvalues and eigenvectors of the discretized LaplaceBeltrami operator and SSBOF can be calculated, then objects are identified, and correlation coefficient means is adopted for the recognition method whose result is better than L-1 similar criterion. Finally, the SSBOF can be contrasted with heat kernel trace and other heat kernel invariant features, also Hu invariant moments. Experimental results show that the recognition rate of spatiallysensitive heat kernel feature SS-BOF is higher than heat kernel invariant features, and compared with the classical Hu invariant moments, the recognition rate is increased.