为提高森林资源监测自动化水平与效率,满足高空间分辨率、大尺度、混合时相及多数据源条件下的森林资源监测需求,采用时空波段集构建与面向对象的多尺度聚类分割方法,对林地占用区域变化进行检测。以陕西省凤翔区为研究区,基于2020年和2021年2期高分一号、高分六号和资源三号3种遥感影像数据,进行波段筛选与特征集构建,通过多尺度区域生长聚类方法将像元级影像聚合为对象单元;在对象级层面综合利用光谱数字量化值与NDVI特征设定阈值,实现林地占用区域的自动检测。结果表明:时空波段集构建与面向对象的多尺度分割方法能够有效分割变化区域和非变化区域,林地占用自动检测精度为82.9%,自动检测得到的地块轮廓与实际占地轮廓之间接近度评价为“较高”以上的结果占75.3%,边界提取准确性和检测效率优势显著。相较于主流的基于像元或面向对象的分类后检测法,时空结合的高维分割方法可有效简化操作流程,在林地占用地块检测中能辅助目视解译,提升对森林资源破坏行为的发现效率。

To enhance the automation level and efficiency of forest resource monitoring,we adopted spatiotemporal waveband set construction to meet the requirements of high spatial resolution,large scale,mixed temporal phases,and multi-source data.Object-oriented multi-scale clustering segmentation was employed to conduct change detection in forest-occupied areas.Taking Fengxiang District in Shaanxi Province as an example,data band screening and waveband set construction for two periods(2020 and 2021)were carried out using three types of remote sensing image data:Gaofen-1,Gaofen-6,and Ziyuan-3.Through multi-scale region-growing clustering,pixel-level images were aggregated into object units.At the object level,a threshold was set using the joint features of spectral digital numbers and NDVI to achieve automatic detection of forest-occupied areas.The results show that the spatiotemporal waveband set construction and multi-scale object-oriented segmentation method can effectively separate changed areas from unchanged areas.The automatic detection accuracy of occupied forest land was 82.9%.75.3% of the automatic detection plot contour and the actual occupied land contour were evaluated as "high",and the edge extraction accuracy and detection efficiency had significant advantages.Compared with the mainstream post-classification detection methods based on pixels or object-oriented approaches,this method effectively simplifies the operation process by using high-dimensional segmentation that combines space and time,and can be applied to detect changes in forest-occupied land plots,assisting visual interpretation,thereby improving the detection efficiency of behavior that damages forest resources.