基于WorldView2和GF-2的面向对象多指标综合植被变化分析

Object-oriented multi-index integrated vegetation change analysis based on WorldView2 and GF-2

【目的】利用高分辨率卫星影像获取精确的植被变化信息对植被资源合理利用及可持续经营有重要意义。传统的基于像元的直接变化检测法容易产生椒盐噪声,而用面向对象分类法结果又严重依赖于分类精度。本文在分析现有算法优劣势基础上,力图找到一种针对高分辨率遥感数据进行植被变化检测的相对客观算法,并验证其有效性。【方法】基于现有的多指标综合变化分析算法(MIICA),提出了面向对象的MIICA。本算法用准确率(P)和查全率(R)分析确定的最优分割参数对前后两期跨传感器影像进行统一分割,利用分割获得的对象影像进行特征参数提取,并用ROC曲线法选择合适的阈值进行变化信息提取并整合,最终获得植被变化位置及方向(植被增多或减少)。【结果】经与基于像元的MIICA及面向对象分类法的比较,本方法的生产者精度高于基于像元的MIICA,用户精度高于面向对象分类法,并且总体精度和Kappa系数分别达到了0.880和0.805。本方法能更好地反映植被变化的位置及形状,也能较准确地检测出一些面积微小的变化。【结论】面向对象的MIICA能弥补基于像元的MIICA和面向对象分类的缺点,提高检测精度,对存在高人为影响的森林公园或自然保护区植被变化分析、植被资源合理利用及可持续经营有重要意义。

[Objective]Using high spatial resolution satellite images to capture accurate information on vegetation change is of great significance for the rational use of vegetation resources and sustainable management.Traditional pixel-based direct change detection methods are easy to cause salt-and-pepper noises and the results of object-oriented classification methods depend heavily on the classification accuracy.After investigating the advantages and weaknesses of the existing algorithms for vegetation change analysis,the major objective of this study was to develop a relatively objective algorithm for vegetation change detection by high spatial resolution remote sensing data,and to verify its effectiveness.[Method]An object-oriented multi-index integrated change analysis(MIICA)algorithm was proposed in the analysis based on a existing MIICA.First,bi-temporal cross-sensor high spatial resolution images were segmented uniformly with the optimal segmentation parameters,which were determined by examining the precision(P)and recall(R)indices,followed by the extraction of feature parameters of the segmented objects.Then the appropriate thresholds objectively determined by ROC(receiver operating characteristic)curves were integrated to derive vegetation change positions and directions(vegetation gain or loss)finally.[Result]Results showed that compared with the pixel-based MIICA and the object-oriented classification method,the producer’s accuracy of our method was higher than that of the pixel-based MIICA,meanwhile the user’s accuracy was higher than that of the object-oriented classification method.And our method had higher overall accuracy and Kappa coefficient,estimated at 0.880 and 0.805,respectively.The detection results of our method could better reflect the positions and shapes of the vegetation change areas,with some subtle vegetation changes clearly detected.[Conclusion]Object-oriented MIICA can improve the shortcomings of pixel-based MIICA and object-oriented classification method,and improve the detection accuracy.The proposed method is essential for the analysis of vegetation changes,the reasonable utilization and sustainable management of vegetation resources in forest parks or natural reserves,where heavy anthropogenic disturbances frequently exist.