SAR image target recognition based on heat kernel

In order to solve low recognition rate which was caused by the speckle noise and deformation in SAR image after earthquake, a new heat kernel feature which possesses the properties of affine invariance and deformation invariance was put forward, and this feature was used in SAR image target recognition. First of all, a generalized kernel fuzzy C-means formula was applied in SAR image segmentation, and target shape in SAR image was extracted. Secondly, the triangle subdivision was obtained by means of Delaunay triangulation formula, and Laplace-Beltrami operator was discretized using cotangent weight scheme, and then heat kernel feature at every point was obtained by making use of the eigenvalues and eigenvectors of the discretized Laplace-Beltrami operator. Thirdly, heat kernel distance between two points was calculated by spectral distance formula, and then distance distribution was used to represent the heat kernel feature in the target shape of SAR image. Finally, L1 similarity criterion was adopted to measure the similarity of two SAR image, recognition result was obtained by comparing similarities of the whole SAR images. Experiments show that, compared with the classical Hu invariant feature which also has affine invariance in images, this method which is based on heat kernel feature shows both higher recognition rate for affine transformation SAR images and deformation SAR images. Consequently, SAR image target recognition method, which is based on the heat kernel feature, is a more effective SAR image recognition method.