基于IR-MAD算法的GF-1影像土地利用变化检测研究

Study on land use change detection of GF-1 image based on IR-MAD algorithm

变化检测是遥感应用领域的研究热点之一,为探究高分一号(GF-1)PMS影像在土地利用变化检测中的适用性和有效性,文章以毕节市金海湖新区办事处为研究区,选取2017、2019两期GF-1 PMS影像为数据源,对比直接分析比较法中的迭代加权多元变化检测(IR-MAD)和基于随机森林(Random Forest, RF)分类后比较法的检测效果。结果表明:IR-MAD算法检测效果良好,可以有效地区分不变区域与变化区域,总体精度和Kappa系数较高,过程不依赖于训练样本的多少,总体精度达到90.78%,Kappa系数为0.86,而随机森林分类算法精度相对较低,检测效果欠佳,总体精度为89.32%,Kappa系数为0.80。因此,IR-MAD算法更适用于小尺度的土地利用变化检测。

Transform detection is one of the research hotspots in the field of remote sensing applications.In order to explore the applicability and effectiveness of GF-1 PMS image in land use change detection.Taking the Jinhaihu New Area Office of Bijie City as the research area,and selecting the GF-1 PMS images of 2017 and 2019 as the data source,the test results of the iterative weighted multiple change detection(IR-MAD)in the direct analysis and comparison method and the post-classification comparison method based on the random forest(RF)were compared.The results show that the IR-MAD algorithm has a good detection effect and can effectively distinguish the constant region from the changing region.The overall accuracy and Kappa coefficient are high,and the process does not depend on the number of training samples.The overall accuracy reaches 90.78%,and the Kappa coefficient is 0.86,while the accuracy of the random forest classification algorithm is relatively low,and the detection effect is poor.The overall accuracy is 89.32%,and the Kappa coefficient is 0.80.Therefore,IR-MAD algorithm is more suitable for small-scale land use change detection.