Research On Crown Extraction Based On UAV Remote Sensing Images And Object Detection

Canopy plays an important role in the overall function of forest ecosystems,and it is also an important information in forest resource surveys.The traditional canopy measurement method is field survey.The measurement results of this method have relatively large errors in specific terrain and forest environments,and the operation is relatively cumbersome.The development of remote sensing technology provides a new method for tree crown extraction.This article summarizes the research status of tree crown extraction at home and abroad,points out the shortcomings of remote sensing data acquisition technology,and explains the current common methods of tree crown extraction.The emergence of drone imaging technology has made up for the shortcomings of remote sensing technology,and the development of deep learning has provided new methods and realization ideas for canopy measurement.The main research content and conclusions of this article are as follows:(1)The UAV remote sensing platform equipped with a camera was used to capture the canopy images of sparsely distributed young pine forest,densely distributed mature pine forest and metasequoia glyptostroboides forest in simple environment.High resolution orthophoto images were generated as data sets,and three object detection algorithms Faster R-CNN,YOLOv3 and SSD models were input for training.(2)The classification indexes and regression indexes were used to analyze and compare the crown recognition and crown width extraction results.It can be obtained:Faster R-CNN and SSD models have excellent canopy detection results in young pine forests,with Accuracy indexes of 99.22%and 96.90%,respectively,Coefficient of Determination(R~2)reached 0.88 and 0.92,respectively;SSD is the best model in crown recognition and crown width extraction of mature pine forest and Metasequoia glyptostroboides forest,the Accuracy indexes are 95.24%and 91.67%,R~2 are 0.94 and 0.88,respectively;(3)The Faster R-CNN model only has excellent results when extracting young pine forests,and the others are poor.In view of this,this study improved the Faster R-CNN model,and compared with the other three models,the detection effect has been greatly improved.In the mature forest and Metasequoia forest,the Accuracy indexes reached 95.74%and 92.92%,respectively,and the R~2 reached 0.85 and 0.84,respectively,which proved the superiority of the improved Faster R-CNN model.The innovation of this study is to verify the feasibility of UAV images and object detection in the field of tree crown extraction through experiments.Compared with the traditional and common measurement and recognition methods,it has the advantages of higher efficiency and lower cost.