Research On Tree Counts Extraction From UAV Imagery Based On Fusion Watershed Algorithm

Investigating and managing forest resources quickly and accurately is one of the important contents of forestry research.With the continuous exploration and development of unmanned aerial vehicle(UAV)and image processing technology,the research on extracting forest structure parameters through UAV imagery has been constantly deepen.This paper used UAV imagery data as the research object to segment imagery,mark and detect individual tree crown and extract the number of trees based on the fusion watershed algorithm.It is necessary to preprocess the UAV imagery before segmenting the imagery and extracting tree counts.The orthophotos were obtained by processing the UAV imagery with Agisoft photoscan software,and the G channel images were selected for subsequent image processing.The images were processed by two methods to extract the number of trees.The first method was selecting the appropriate structural elements based on the characteristics of the G channel images,and building morphological filtering and removing noise of the imagery through morphological hybrid open and close operation,and the corresponding binary images were obtained by local maximum calculation.Then,these images were processed by the watershed segmentation algorithm which based on distance transform,to segment the images,mark the centroid of the tree crown and extract the number of trees.But,in this method,the removal efficiency of the noise needed to be further improved.In order to better extract tree counts and improve the detection and extraction accuracy,the paper proposed another new method which fused the watershed segmentation algorithm with the Mean Shift algorithm to extract the number of trees.The G channel images were smoothed by the Mean Shift cluster algorithm.Then,the smoothed images were converted into the binary image,and were fed into the method which combining morphological hybrid opening and closing reconstruction operation and the watershed segmentation algorithm which based on Euclidean distance transformation to mark the centroid of the tree crown and extract the tree counts.The average extraction accuracy of the two methods in this paper were above 90%,when the result was compared to the ten manually marked and counted plots,and the over segmentation problem of the traditional watershed algorithm was effectively improved.It provides an effective method for forest resource investigation and subsequent acquisition of relevant forest structure parameters.