LAUSR

Abstract Aiming at the requirements of effectively recognizing ships using optical remote sensing data, a ship recognition method of probability generative model based on variational inference is proposed. Firstly, according to the principle of region partition by pixel gray level, the feature extraction based on local spatial gray information of pixels is built, which can efficiently measure similarity degree between the current pixel and its neighborhood structure in a search window, and can be able to provide ship recognition feature stably and accurately. In addition, theoretical analysis shows that the feature extraction method has greatly reduced time complexity of the algorithm. Secondly, a classification method of probability generative model based on variational inference is presented, in this model, manifold similarity based on local reverse entropy operator is used to measure the discrepancy of samples, and the neighborhood samples are selected from top of defined dominant set. Finally, experimental results on real data demonstrate that the proposed method can be obtained a higher recognition performance than k-nearest-neighbor (KNN), support vector machine (SVM), hierarchical discriminant regression (HDR), probability generative model (PGM), dynamic probability generative model (DPGM), it satisfies the time efficiency requirements of ship recognition in projects.