Abstract

The ship extraction from remotely sensed images has attracted much attention. It can supervise fisheries and manage marine traffics to ensure its safety. With the development of satellite and intelligence, automatic ship extraction has replaced the traditional manual ship detection. Ship extraction methods can be divided into two groups by image sources: synthetic aperture radar (SAR) image based and optical image-based methods. The SAR image-based ones have advantages of all-weather time and big difference of ships and sea; therefore, it has extensively been studied. A constant false-alarm rate (CFAR) detector is a usually used ship extraction algorithm, which assumes a certain background distribution for SAR images, such as k-distribution, Gamma distribution, Gauss distribution, and other combination. However, the ship extraction from SAR images also has its limitations, such as low resolution of SAR images, relatively long revisit cycle, and the complicated sea clutter. Optical remotely sensed images have the advantage of high resolution and relatively short revisit cycles, and have more detailed texture, spectral and shape information. Therefore, the ship extraction from optical remotely sensed images were widely studied. Shape features and gray intensity are two mainly used optical image-based ship extraction methods. Shape features based methods use shape and edge information to extract ships. For example, most ships have narrow bow area and parallel hull edges, which are easily detected because of the big difference between ships and water. Many researchers have adopted this idea to detect ships. Due to the big gray intensity distinction between ships, and water, gray intensity was exploited to detect ships by segmenting images. With the development of artificial intelligence, neural network has been widely studied in much research, especially deep learning, and deep neural network.

Description