Abstract: It is difficult for the visible image of unmanned aerial vehicle (UAV) to recognize some confused backgrounds and targets. To solve this problem, the paper collected the pitaya images matching point cloud data in the areas of Karst Valley through using the four-rotor drones. After processing the original point cloud data with desiccation, filtering, normalization and so on, through the establishment of the digital elevation model (DEM), digital surface model (DSM) with high accuracy, and then a high-precision canopy height model (CHM) was established. Furthermore, with the visual interpretational strains of pitaya as a reference, the definite numbers of pitaya was identified, extracted and verified. The results indicated that the influence of weeds under the plants could be eliminated to some extent, through using UAV images matching point cloud data and canopy Height model. If the height of infrastructure or ground object in the sample area was approach to that of the pitaya, it would lead to false extraction. The highest false extraction rate was 8.55%, and the highest missed extraction rate was 12.28%. Among all the sample areas, the precision of the number of pitaya plants which were extracted by the seed points was more than 92.38%; yet that of pitaya plants extracted by using the vegetation canopy was more than 90.68%. It indicated that the means of using UAV image matching point cloud data to extract the pitaya, had the characteristics of high speed, easy to deposit, low cost, and reliable accuracy, which was suitable for the rapid extraction of the number of crop in Karst mountain areas. It could mutually complement with the extraction method based on the color index.

Key words: unmanned aerial vehicle, image matching point cloud, canopy height model, single plant extraction, pitaya, Karst Canyon Area