滑坡遥感影像的傅里叶描述子轮廓特征提取

Fourier Descriptor Contour Feature Extraction from Landslide Remote Sensing Image

针对遥感影像滑坡的轮廓不受地域、滑坡种类、纹理和颜色特征影响的特点,且同时具有很好的旋转及尺度不变性,是滑坡遥感影像中较为稳定的一项特征,提出了一种基于傅里叶描述子进行滑坡遥感影像轮廓特征提取的方法。创建了一个小型滑坡遥感影像样本库,在这些滑坡影像的基础上,通过傅里叶描述子低频特性选取少量的描述子对滑坡轮廓进行重构;利用Hu不变矩验证了滑坡轮廓平移、旋转和尺度的不变性;以多分类支持向量机进行分类。通过验证表明,傅里叶描述子可以有效地提取滑坡的轮廓特征,识别效果优于Hu不变矩和融合特征(傅里叶描述子和Hu不变矩)。

In view of the fact that the contour of landslide in remote sensing images is not affected by region,landslide type,texture and color characteristics,and has good rotation and scale invariance at the same time,it is considered a relatively stable feature in the remote sensing image of landslide.A method for contour feature extraction of landslide remote sensing images based on Fourier descriptors is proposed.Firstly,a small sample database of landslide remote sensing images is created.Based on these landslide images,a small number of descriptors are selected through the low-frequency characteristics of Fourier descriptors to reconstruct the landslide contour,and then the invariance of translation,rotation and scale of landslide contour is verified by Hu invariant moment.Finally,multi-classification support vector machine is used for classification.The verification shows that the Fourier descriptors can effectively extract the contour features of landslide,and the recognition effect is better than Hu invariant moment and fusion feature(Fourier descriptor and Hu invariant moment).