纹理与几何特征信息在高空间分辨率遥感影像分类中的应用

Application of Texture Features and Geometric Feature Information in High Spatial Resolution Remote Sensing Image Classification

为充分利用高空间分辨率遥感影像所包含的纹理与几何特征信息,提高影像的分类精度,应用面向对象技术和Log-Gabor小波来提取影像纹理和几何特征信息,并结合影像光谱信息对影像进行分类。选用的分类器为粒群优化算法优化参数的支持向量机分类器,使用资源三号和高分二号两种不同传感器的卫星影像对方法进行验证,影像数据地物类型包含林地、草地、裸地、沙地、建筑物和道路等典型实体对象。通过对实验结果的目视比较和统计分析结果表明,所提方法分类结果精度较传统方法有所提高。

In order to make full use of the texture and geometric feature information contained in the high spatial resolution remote sensing image to improve the classification accuracy of the image, object-oriented techniques and Log-Gabor wavelet are used to extract the image texture and geometric feature information combined with spectral information.The selected classifier is SVM classifier with particle swarm optimization algorithm. The method is validated by using satellite images of ZY3 and GF2 sensors. The image data types include woodland, grassland, bare land, sandy land, buildings and roads and other typical entity objects.Through the visual comparison and statistical analysis of experimental results, it shows that the accuracy of the method proposed in this paper is higher than the traditional method.