针对无人机遥感技术在提取单株立木信息的限制性问题，提出一种新的自动单株立木信息提取方法。对原始无人机影像进行光谱信息增强处理以突出局部细节特征；通过引入DBI指数自动化确定K-means聚类方法的最优聚类数目，进而对影像像素进行标记；通过利用高斯马尔可夫随机场模型进一步对影像进行分割；使用数学形态学算子等方法对分割结果进行后处理得到单株立木树冠信息，通过图像几何矩原理计算得到单株立木位置以作为其识别的依据。结果表明，应用该提取方法，油松林区和樟子松林区单株立木识别总体精度分别为89.52%和95.65%、单木树冠提取精度分别为81.90%和95.65%，均具有较好地适用性。该方法不需要大量的人工干预和先验知识的输入，大大提高提取方法的自动化程度。

Extraction of individual tree information is significant in managing forestry resources and protecting the ecological environment. In addition, this strategy has been extensively applied to forests using high spatial resolution satellite images or Lidar data in recent years. The rapid development of unmanned aerial vehicle remote sensing makes itself play an important role in this field.
This study uses high spatial resolution aerial remote sensing images to extract individual tree information automatically. The individual tree information includes individual tree extraction and crown delineation.
A novel extraction method of individual tree information was proposed in this study. First, the spectral information of the original images was enhanced through decorrelation stretch. Spectral information enhancement aimed to expand the coupling degree of image information by stretching the principal component information of the bands that are correlated, thus increasing the color saturation of the image. Second, the optimal cluster number of K-means clustering algorithm was obtained by introducing a DBI index, then the image pixels were marked based on the K-means clustering algorithm. On this basis, a GMRF model was constructed for image segmentation. Finally, the segmentation results were post-processed through mathematical morphology to complete individual tree crown delineation, and individual tree detection was based on the position of individual tree calculated by image geometric moment. This study acquired the reference data through visual interpretation to evaluate the result.
The verification result showed that the overall accuracies of individual tree detection through this method are 89.52% and 95.65% for medium and low canopy density forests, correspondingly. The accuracies of individual tree crown delineation achieved 81.90% and 95.65% for medium and low canopy density forests, respectively. Therefore, the proposed method is better than the object-oriented method in terms of extraction effect through comparison and analysis.
The proposed method can be used to obtain the individual tree information of forestry management rapidly and has high extracting precision based on the verification of results. The proposed method does not require considerable manual intervention and prior knowledge, thus significantly improving the degrees of automation.