论文部分内容阅读
对刺槐林健康状况进行准确分类制图,是进行刺槐林健康状况评估与生态修复的前提。以高分辨率IKONOS影像、基于影像提取的不同窗口、不同灰度共生矩阵纹理信息以及反映局部空间自相关的Local Getis-Ord Gi(Getis统计量)为数据源,结合实测生态样方数据,利用多决策树的组合分类模型随机森林(RF)对刺槐林健康进行分级,对6种方法的分类精度进行了比较且对分类变量的重要性进行了排序。结果显示:19m×19m是最佳纹理计算窗口;灰度共生矩阵均值是最优纹理变量;基于波段4计算的Getis统计量对RF分类具有最重要的作用;较之利用全部光谱、纹理和Getis统计量的80个波段/变量,利用前向选择得到的前16个重要性变量进行RF分类,获得了最高的分类精度(总精度为93.14%,Kappa系数为0.894)。研究证实了从高分影像提取的空间特征信息有助于提高对具有规则分布格局的人工刺槐林健康等级的分类精度;前向选择方法可以利用较少的预测变量获得较高的分类精度。
Accurate classification mapping of the health status of Robinia pseudoacacia forest is the prerequisite for evaluating the health status of Robinia pseudoacacia forest and ecological restoration. Based on the high-resolution IKONOS images, the local window based on image extraction, texture information of different gray level co-occurrence matrices and Local Getis-Ord Gi (Getis statistics) reflecting the local spatial autocorrelation, The combined classification model of multiple decision trees Random Forest (RF) stratified the health of Robinia pseudoacacia forests. The classification accuracy of the six methods was compared and the importance of categorical variables was ranked. The results show that: 19m × 19m is the best window for texture calculation; mean value of gray level co-occurrence matrix is the best texture variable; Getis statistics based on band 4 have the most important effect on RF classification; compared with all spectra, texture and Getis The 80 bands / variables of the statistics were used for RF classification using the first 16 importance variables obtained from forward selection, resulting in the highest classification accuracy (overall accuracy of 93.14% and Kappa coefficient of 0.894). The research confirms that the spatial feature information extracted from high-resolution image helps to improve the classification accuracy of artificial Robinia pseudoacacia forest with a regular distribution pattern. Forward selection method can use fewer predictors to obtain higher classification accuracy.