基于纹理信息的森林类型遥感识别技术

Remote Sensing Identification on Forest Types Based on Texture Information

为提高大区域TM影像对针阔混交林的识别精度,充分考虑遥感影像像元值的随机性和空间性,以盘古林场有林地TM遥感影像为例,结合地统计学知识,利用变异函数计算图像纹理信息,分析了影像纹理信息提取的重要因子,确定选取绝对变差函数为计算方法,以9×9像元为窗口,4像元为步长,计算方向为全方向对盘古林场有林地部分提取纹理信息并与原始光谱信息及归一化植被指数相结合,采用经典分类器最大似然法对影像进行分类。结果表明,辅以纹理信息的最大似然法分类精度为85.333 3%,Kappa指数为0.78,达到了区别针阔混交林的目的。

The research was conducted to improve the classification precision of coniferous, broadleaf and conifer-broadleaf mixed forest by TM images for large areas by using geostatistics considering the randomness and spatial distribution of the pixel values of the TM data. Taking the Pangu forest farm as the research region, the texture information of TM images were calculated using semi-covariance functions with a window of 9×9 pixels and 4 pixels step size for all directions. The absolute semi-covariance was selected through analyzing the factors affecting the texture information. The forest classification results were obtained by the maximum likelihood method with the texture information in the combination of the original spectral information and normalized differential vegetation index （NDVI）. The accuracy of identifying forest type by the maximum likelihood method with the texture information is 85. 333 3% , and the Kappa coefficient is 0.78.