基于变化检测-CART决策树模式自动识别沙漠化信息

Automatic Recognition of Desertification Information Based on the Pattern of Change Detection-CART Decision Tree

目前沙漠化遥感监测存在目视解译的局限性、数据源的约束性、遥感信息利用率低等问题。基于此,以民勤盆地为试验区,首先采用图像差值、最大值合成及二维最大类间方差等方法,检测1994年、2014年两期Landsat图像的变化像元,然后利用分类与回归树(CART)算法构建决策树,自动提取了2014年沙地信息,最后将变化检测结果与沙地信息进行空间叠置分析,并实现了沙漠化信息自动识别模式。研究表明,变化检测-CART决策树模式精度为89.43%~93.00%,在95%置信水平上其置信区间介于85.90%~98.00%,显然其精度具有较高可信度;该模式不仅能够充分利用丰富遥感信息而且可排除多余信息的干扰。可见,变化检测-CART决策树模式是识别沙漠化信息的有效方法之一,将对沙漠化防治工程具有重要应用价值。

The desertification of remote sensing monitoring has some problems, such as visual interpretation limitation, constraint of data source and low utilization rate of remote sensing information. Based on this, taking Minqin basin as the test area, firstly, this paper detects the change pixel of two Landsat images in 1994 and 2014, by employing the methods of image difference, maximum value synthesis and two-dimensional maximum between- class variance. Secondly, the decision tree is constructed by the classification and regression tree （CART） algo- rithm then automatically extracts the sandy land information of 2014. Finally, spatial overlay analyze the results of change detection with sandy land information, and realize the pattern of automatic recognition on desertification in- formation. The research shows that the accuracy of change detection-CART decision tree pattern is 89.43% to 93%, and the confidence interval is between 85.90% and 98% at 95% confidence level, clearly that the reliability of its accuracy is relatively high. This pattern not only can make full use of the abundant remote sensing informa- tion but also can exclude the interference of redundant information. Obviously, the change detection-CART decision tree pattern is one of the effective methods to identify the desertification information, and it will have important ap- plication value to the desertification control project.