Research On Standing Tree Attributes Measurement System Based On Android Platform

With the development of digital forestry,the measurement accuracy and efficiency requirements of standing tree factors in forestry resource surveys are getting higher and higher.Traditional measurement tools generally have the problems of inconvenient carrying,low measurement efficiency and high cost.Although the existing mobile-based standing tree factor measurement method solves the problems in the traditional method,the measurement still requires the assistance of external equipment.In response to these problems,this thesis proposes a method of segmentation of standing tree image suitable for mobile terminal,and builds a measuring model of standing tree factor based on photogrammetry principle,and develops a measuring system of standing tree factor based on Android platform,realizing fast and convenient measurement of diameter at breast height and tree height.The main tasks of this thesis are as follows :(1)aiming at the problem that MeanShift algorithm has a large amount of computation and it is difficult for mobile terminal to support its operation,this algorithm is improved by color space transformation.(2)The Mask R-CNN network with improved Res Net network depth is used to deploy the trained tree image segmentation model to achieve real-time online calling.(3)Fast self-calibration of camera is realized by using Zhang Zhengyou calibration method in mobile terminal.(4)Based on the principle of photogrammetry and trigonometry,the measurement model of depth,DBH and height of standing trees is constructed,which can realize the rapid calculation of depth,DBH and height of trees.(5)Developed the standing tree factor measurement system and realized the standing tree factor measurement based on mobile terminal.The improved MeanShift algorithm in this thesis reduces the processing time by 80%.The recall rate of MeanShift’s Grab Cut standing-tree segmentation method increases 9.5% and 6.7% respectively compared with the accuracy before improvement.The recall rate and accuracy of the method based on Mask R-CNN training were 97% and 98%,respectively.The average relative error of the height measurement model is3.46%,and that of the DBH measurement model is 2.35%.The developed measuring system of standing tree factor does not need auxiliary equipment and reference objects,which reduces the measuring requirements and improves the measuring efficiency of standing tree factor.The innovations of this thesis are as follows :(1)using smart phones as measuring devices,the cost of forest resource survey is reduced,and data processing in the field is avoided,and the height and DBH of standing trees are quickly measured.(2)Explore the segmentation method of standing wood image based on Grab Cut to improve the segmentation accuracy,and improve the Mask R-CNN network to achieve high-precision segmentation of standing wood image.(3)Make full use of the built-in orientation sensor of the mobile phone to realize the rapid extraction of the depth of the standing image without reference.