A Level Set Based Method for Land Masking in Ship Detection Using SAR Images

This paper presents an efficient approach to obtain land masking using synthetic aperture radar (SAR) images, which is based on level set method and fully convolutional network (FCN) classification. First, the level set method is applied to the cropped SAR input image for initial contours. Second, FCN model has been trained to classify the input image by two labels of water and land which will find the possible region existing real coastlines, named region of interest (RIO). Third, to select the desired contour from extracted boundaries in the first step, according to the proportions to be covered in RIO of step two. Then, final land masking can be obtained after morphological processing and color filling. The method proposed in this paper is fast and accurate enough for ship detection in high-resolution SAR images. It is also robust to speckle noise and geographical changes. Experimental results on GF-3 SAR images show good performance of this method.