Single Tree Segmentation in Close-Planting Orchard Using UAV Digital Image

  Objectives  Single tree canopy extraction is of great significance for fruit tree health status, nutritional composition, and yield prediction.As a low-cost, low-risk data source, high-resolution remote sensing images obtained by drones provide new technical means for accurately estimating tree numbers and delineating tree canopy profiles. Using unmaned aerial vehicle(UAV) digital images as the data source, a single-tree segmentation algorithm based on regional seed blocks is proposed to solve the single-tree segmentation problem of densely planted fruit trees.

  Methods  The canopy of the fruit tree is extracted by the maximum likelihood method to generate digital surface model (DSM)of the canopy, and Gaussian filter is combined with morphological opening and self-adapted threshold segmentation generates regional seed blocks as the basis for tree statistics and as the marker of the marked-controlled watered segmentation.

  Results  The results show that the overall tree recall rate is 95.22%, the precision rate is 99.09%. The overall accuracy rate of single tree contour extraction is 93.45%, the overall omission error is 5.87%, and the overall commission error is 0.90%. Compared with the previous local maximum seed point extraction results, the overall accuracy is 18.66% higher, and the precision of the fine crown contour extraction is 17.75% higher.

  Conclusions  The watershed method based on the regional seed block can effectively prevent the overlapping area of the canopy from being repeatedly divided into multiple fruit trees. On the basis of preserving the canopy outline of the fruit tree to the greatest extent, the segmentation error of the fruit tree is reduced.It provides a reference for the method of extracting the crown of a single fruit tree in a densely planted orchard in flat terrain.