针对复杂背景下单时相极化合成孔径雷达(PolSAR)图像滑坡泥石流毁损区检测正确率低、虚警多的问题,提出一种联合支持向量机(SVM)与层次分析法(AHP)的滑坡泥石流毁损区自动检测方法。首先,为充分利用极化信息,基于相干矩阵的九个元素构建特征向量并进行SVM 地物分类,初步提取出区域分类结果;其次,基于归一化极化散射功率、同极化 相关系数和雷达植被指数构建因素集合,并采用AHP 方法优化评价因素权重,改善评分标准,更准确地判定滑坡泥石流毁损区的表面散射特性;最后,通过逻辑“与”运算将二者结果结合起来,实现了准确检测滑坡泥石流毁损区域并有效降低虚警的目的。利用2015 年深圳地区和2018 年日本北海道伊布里地区的ALOS-2/ PALSAR-2 数据进行了滑坡泥石流毁损区检测实验。实验结果表明,所提方法相比现有方法显著抑制林地、耕地等目标的虚警率,且在这两个滑坡案例中的滑坡目标检测正确率分别提升了约6%和2%。

Aiming at the problems of low accuracy and many false alarms of landslide detection in single-phase polarimetric synthetic aperture radar (PolSAR) image with complex background, a landslide detection method combining support vector machine (SVM) and analytic hierarchy process (AHP) is proposed in this paper. Firstly, in order to make full use of the polarization information,the nine elements of the polarization coherence matrix are used as the feature set of SVM to classify landslide features,and the landslide area is preliminarily extracted. Secondly, the parameter set is constructed based on normalized polarization scattering power, co-polarization correlation coefficient and radar vegetation index, and the weight of evaluation parameters is optimized by AHP analysis method, the scoring standard is improved, so as to determine the surface scattering characteristics of landslide area more accurately. Finally, the results of SVM and AHP are respectively performed by morphological operation, and then the logical operation is performed to obtain the final landslide detection results. The purpose of accurately detecting landslide area and effectively reducing false alarm is realized. In this paper, the ALOS-2/ PALSAR-2 data of Shenzhen landslide in 2015 and Hokkaido Iburi in 2018 are considered to verify the performance of the proposed method. Compared with the existing methods, the experimental results indicate that the method significantly inhibits the false alarm of forest, fields and other targets, and the accuracy of landslide detection in these two landslide cases is improved by about 6% and 2% respectively.

TN957. 52

国家自然科学基金面上资助项目(41674032)