Ship Detection Based on Superpixel-Level Hybrid Non-local MRF for SAR Imagery

Abstract

Speckle noise interference and inshore ship detection have remained difficult problems associated with the detection of ships in SAR images. The segmentation method based on the Markov random field (MRF) model can classify the pixels of an image for target detection. However, the traditional MRF method is not robust against speckle noise and cannot use the high-level information of the image. To overcome these problems, this paper proposes a hybrid non-local MRF model based on superpixel segmentation for the detection of ship targets at sea and in dock areas in SAR images. Superpixel segmentation can segment an SAR image into blocks of pixels with similar attributes to suppress the effect of coherent speckle noise in the image. In addition, a hybrid non-local energy function is proposed, and it can comprehensively consider local and non-local information and greatly improve the accuracy of image segmentation. The experimental results show that the proposed method can improve the detection accuracy and significantly reduce the false alarm rate compared with the those of the traditional MRF method.