整合无人机和面向对象的农村居住环境信息提取

无人机遥感和面向对象图像分析技术在环境监测中得到越来越多的发展。然而在科学文献领域，使用无人机和面向对象制图农村居住环境的文献仍然很少。因此本研究构造一个整合框架用于提取农村居住环境中各类地物信息。首先利用尺度参数评估(ESP, Estimation of Scale Parameter)工具和专家判断来确定最优分割尺度参数；然后分别采用专家规则集和监督分类算法提取农村居住环境中各类地物；最后采用基于面的精度评价方法对分类性能进行评估。结果表明，利用ESP工具和专家判断确定最优分割尺度是可行的。总体精度为75.19%，说明基于规则的提取方法对研究区各类地物的提取效果不佳。但在农村居住环境中利用模板匹配结合阈值规则对太阳能热水器提取精度达92%。分析训练样本和特征对随机森林(RF，Random Forest)、支持向量机(SVM, Support Vector Machines)和K最近邻 (KNN, K-Nearest Neighbor) 分类器分类结果的影响，说明RF分类器对农村居住环境分类效果最好，总体分类精度高达91.34%。研究结果表明:该框架在农村居住环境地物提取方面是一种有价值的工具。

Integrating UAV and Object-based Image Analysis for Rural Residential Environment Information Extraction

The remote sensing of Unmanned Aerial Vehicle (UAV) and Object-Based Image Analysis（OBIA） technologies have advanced increasingly for environmental monitoring in recent years. However, references to the uses of UAV and OBIA for mapping rural residential environment are still scarce in the field of scientific literature. In this study, an integration framework was developed to extract various ground objects in rural residential environment. First, Estimation of Scale Parameter (ESP) tool and expert judgement were used to identify the optimal Segmentation Scale Parameter (SSP). Then, the expert rule-sets and supervised classification algorithms were applied to extract information of ground objects in rural residential environment, respectively. Finally, the performance accuracy was evaluated by using an area-based method. The results indicated that using ESP tool and expert judgment to determine the optimal SSP is feasible. Furthermore, the overall accuracy (OA) is 75.19%, indicating that the rule-based extraction method is not good at extracting all kinds of ground objects in study area. However, Solar Water Heaters (SWHs) were successfully extracted in rural residential environment by using template matching combined with threshold rules and the extraction accuracy can achieve 92%. Moreover, the influences of training samples and features were analyzed on the classification results of the Random Forest (RF), Support Vector Machines (SVM) and K-Nearest Neighbor (KNN) classifiers, showing that the RF classifier has the best classification result, with its value of OA reaching 91.34%. The results indicated that the integrate framework is a valuable tool in the extraction of ground objects from rural residential environment.