Target recognition in ISAR images based on relative phases of complex wavelet coefficients and sparse classification

In this paper, we present a novel approach for radar automatic target recognition on inverse synthetic aperture radar (ISAR). This approach is composed by two complementary steps: feature extraction and recognition. For the feature extraction step, we adopt a statistical modeling of the ISAR image in the complex wavelet domain. For doing so, we apply the dual-tree complex wavelet transform (DT-CWT) for each ISAR image in the database. After that, the relative phases of the resulting complex coefficients are computed. These relative phases are after statistically modeled using the Von Mises distribution. The estimated statistical parameters compose the feature vector of the ISAR images. Regarding to the recognition rate, the constructed feature vector is fed into the sparse representation based classification (SRC). More precisely, the training feature vectors are used as the atoms of a dictionary. The test feature vector is recognized according to its sparse linear combination with the dictionary. Extensive experiments and comparisons with other methods on ISAR images database demonstrate the effectiveness of the proposed approach.