Abstract: In order to make full use of multi-band spectral features of remote sensing images and improve the accuracy of extracted waterline, waterline extraction model from remote sensing data based on Kernel Principal Component Analysis（KPCA） and spectral features constrained is presented. First, the KPCA method is used to get the spectral features of water training samples, the maximum likelihood method is used to estimate the parameters of probability density function of water in feature space and then spectral feature term is constructed. Then image data term is constructed based on Geodesic Active Contour（GAC） model and combined with term spectral feature term and the image data term to establish waterline extraction model. At last, the experiments on Landsat TM datasets validate the effectiveness of the proposed model. Compared with Geodesic Active Contour（GAC） model and Distance Regularized Level Set Evolution（DRLSE） model, this model can improve the accuracy of extracted waterline in ensuring a certain speed of operation.

Key words: Kernel Principal Component Analysis（KPCA）, spectral feature, Geodesic Active Contour（GAC）, Distance Regularized Level Set Evolution（DRLSE）, waterline

摘要： 为充分利用遥感影像的多波段光谱特征，提高水边线的提取精度，提出了基于核主元分析（KPCA）光谱特征约束的水边线提取模型。利用KPCA变换提取水体样本的光谱特征，采用最大似然法估计特征空间中水体光谱特征概率密度函数的特征参数，进而构建水体的光谱特征项。以测地线活动轮廓（GAC）模型为基础，建立图像数据项。结合光谱特征项和图像数据项建立水边线提取模型。在Landsat TM数据集上进行的水边线提取实验验证了算法的有效性，与GAC模型和基于距离正则化的水平集方法（DRLSE）相比较，该算法提取的水边线，在保证一定运行速度的情况下，更准确。

关键词: 核主元分析（KPCA）, 光谱特征, 测地线活动轮廓（GAC）, 基于距离正则化的水平集方法（DRLSE）, 水边线