Abstract: Island is an important carrier of ocean exploitation and utilization. Research on the rapid segmentation of island boundary has important scientific value and strategic significance. Remote sensing technology supplies important data source for island research due to its characteristics of far distance earth observation. In this paper, considering the characteristics of remote sensing data, such as multi-bands, strong real-time, and large area coverage, an improved model is proposed for island boundary rapid segmentation. The proposed model has two aspects: the coarse segmentation of island boundary is done based on K-means algorithm; and then taking the result of coarse segmentation as input data, the re-segmentation of island boundary is optimized based on the improved level set segmentation algorithm. Finally, taking the boundary segmentation of two islands as an example, which are located on Fuzhou, the improved segmentation model is compared with Chan-Vese model, geodesic active contour model and selective binary and Gaussian filtering regularized level set model. The result shows that, the consuming time and the iterations of the proposed model are reduced at least 50% than other models; the accuracy of the results based on the proposed model is closer to the value segmented by visual interpretation. In short, the proposed model overcomes the low efficiency and over-segmentation of the traditional segmentation methods, and presents a rapid method for island census.

Key words: remote sensing data, level set segmentation algorithm, island boundary

摘要： 海岛是海洋开发和利用的一个重要载体。海岛边界的快速分割方法研究具有重要的科学价值和战略意义。遥感技术由于具有远距离对地观测的特点，为研究海岛提供了重要的数据资源。鉴于遥感数据波段数多、实时性强和面积覆盖广等特点，提出了一种海岛边界快速分割模型。该模型分为两方面：基于K均值聚类算法实现海岛边界的粗分割；将海岛边界的粗分割结果作为输入，基于水平集方法实现海岛边界优化。以福建省福州海域某两个海岛的边界分割为例，将改进模型与传统的Chan-Vese模型、测地活动轮廓模型和二值化高斯滤波水平集模型进行比较，结果表明：改进模型的计算耗时和迭代次数较传统分割模型至少降低了50%；改进模型的海岛边界结果精度更加逼近目视解译分割结果。由此可见，改进模型有效地解决了传统分割模型效率低、过分割等现象，为海岛普查等提供了一种快速分割方法。

关键词: 遥感数据, 水平集分割算法, 海岛边界