基于均值漂移的遥感图像海陆边界分割算法

A sea-land segmentation algorithm for remote sensing images based on Mean-Shift

光学遥感图像海陆边界分割是海洋近岸目标检测和识别的重要技术基础。由于光学遥感图像分辨率的提高,海陆边界分割存在背景复杂、干扰多等问题。为解决复杂背景下高清遥感图像海陆边界分割问题,提出了一种新的海陆边界分割算法。该方法包含三个部分:使用均值漂移算法将遥感图像处理成若干同质区域;采用一种新的基于扫描线的方法选择海水区域种子像素点,利用区域增长算法填充海水区域;通过连通区域分析的方法分离陆地部分,得到海陆分割结果。实验证明,该方法能够对于复杂背景下的光学遥感图像实现准确、稳定的海陆分割,算法具有较强的鲁棒性和准确性。

Sea-land segmentation of optical remote sensing images is one of the key technologies for near shore targets detection and recognition.With the development of optical remote sensing,the resolution of the optical remote sensing images is higher.A new method based on Mean-Shift（MS）is proposed to solve the sea-land segmentation problems in the complex scenes.MS algorithm is used to separate the image into several homogeneous regions.A new method based on scan line is adopted to choose the seed pixel of the water area,and with the seed pixel.The region growth method is used to fill the water area.The sea-land segmentation template is extracted through analyzing the connected regions.Experiment results demonstrate that the proposed method can deal with the optical remote sensing images with complex scenes,and also show the accuracy and robustness.