目的 应用高分辨率遥感影像快速准确提取单木树冠信息，对现代森林管理具有重要意义。面向对象的多尺度分割方法能有效地解决基于像元特征分析的局限，是单木树冠提取的重要技术途径。本文对比分析了不同遥感平台和人工林树种的树冠提取精度，探究实验方法针对不同尺度影像数据和树种的优势及适用性，并结合调查目的为影像数据的选取提供参考。方法 以广西壮族自治区高峰林场为研究区，选取低空无人机CCD、机载CCD和星载高分二号遥感影像数据，针对树冠区域与背景区域的对比度效果不佳的问题，首先采用小波变换进行图像增强处理，去除影像噪声，增强树冠与背景的对比度；然后应用面向对象的多尺度分割方法，排除背景区域的干扰，针对树冠区域进行单木树冠的快速提取；最后对3种影像下提取的杉木和桉树人工林单木树冠的流程和方法，以及树冠提取精度进行研究分析。结果 采用小波变换对无人机和机载平台影像增强效果显著，无人机平台下桉树和杉木实验区单木分割精度分别为87%和93.3%，冠幅估测精度为84.2%和85.1%；机载平台下桉树和杉木实验区单木分割精度为89%和91.1%，冠幅估测精度为83.9%和84.4%；而小波变换对星载平台影像增强效果不佳，桉树和杉木实验区的单木分割精度为82%和89%，冠幅估测精度为72.3%和73.3%。结论 在无人机和机载平台下，应用多尺度分割得到的树冠提取精度相接近；在星载平台下，直接应用多尺度分割进行单木树冠提取，受影像自身空间分辨率的局限，提取精度低于前两种平台，但也能够满足森林调查的基本需求。
Objective This method applies high resolution remote sensing image to extract individual tree crown information quickly and accurately, which can have important significance for modern forest management. Object-oriented multi-scale image segmentation method can effectively solve the limitations of pixel feature analysis and is an important technical approach to individual tree crown extraction. This paper compares and analyzes the tree crown segmentation accuracy of different remote sensing platforms and artificial forest species, explores the advantages and applicability of the experimental methods for different scale image data and tree species, and provids reference for the assortment of image data combined with the purpose of investigation.Method Taking Gaofeng Forest Farm of Guangxi Zhuang Autonomous Region as the research area, the UAV CCD, airborne CCD and spaceborne GF-2 remote sensing image data were selected. Aiming at the poor contrast effect between the crown area and the background area, the image enhancement processing was firstly performed by wavelet transform to remove the image noise, enhance the contrast between the crown and the background, and then apply the object-oriented multi-scale segmentation method to eliminate the interference of the background area. Rapid extraction of single tree crown for canopy areas was taken. Finally, the process and method of extracting single tree crown of Eucalyptus robusta and Cunninghamia lanceolata plantation under three kinds of images, and the accuracy of crown extraction were studied and analyzed.Result Wavelet transform is effective in enhancing UAV and airborne images. The individual tree crown segmentation accuracy of Eucalyptus robusta and Cunninghamia lanceolate in UAV platform was 87%, 93.3%, with tree crown estimation accuracy of 84.2%, 85.1%, respectively. The individual tree crown segmentation accuracy of Eucalyptus robusta and Cunninghamia lanceolate in airborne platform was 89%, 91.1%, with tree crown estimation accuracy of 83.9%, 84.4%, respectively. However, wavelet transform is not appropriate for image enhancement of spaceborne platform. The crown segmentation accuracy of Eucalyptus robusta and Cunninghamia lanceolata in spaceborne platform was 82%, 89%, with tree crown estimation accuracy of 72.3%, 73.3%, respectively.Conclusion In UAV and airborne platform, the precision of tree crown extraction by multi-scale segmentation is close. In spaceborne platform, the extraction accuracy of the individual tree crown is lower than that of the former two platforms because of the limitation of spatial resolution of the image, and the direct application of multi-scale segmentation to single tree crown extraction. But it can also meet the basic needs of forest survey.