基于无人机可见光影像的毛竹林郁闭度估测方法

  目的  由于毛竹Phyllostachys edulis的生长特点和经营特点，使得毛竹林郁闭度在毛竹林经营中尤为重要，只有保持适宜毛竹生长的郁闭度，才能提高毛竹生产力。研究无人机可见光影像的毛竹林郁闭度估测方法，可实现实时快速获取毛竹林的郁闭度。  方法  以普通旋翼无人机可见光毛竹林影像为研究对象，基于像元的阈值分类、像元的监督分类、多尺度分割的阈值分类、多尺度分割的监督分类等4种方法，选取不同钩梢和郁闭度的样地36个，利用现有软件和MATLAB编程，对各样地的毛竹林竹冠区域进行快速提取，进而估算林分郁闭度，对比目视解译的郁闭度真值计算各方法的估算精度，利用单因素方差分析比较4种方法在不同钩梢和不同郁闭度下估算郁闭度的表现。  结果  基于像元的阈值分类、基于像元的监督分类、基于多尺度分割的阈值分类、基于多尺度分割的监督分类等4种方法的郁闭度估算总体精度依次为91.81%、92.96%、93.47%、98.86%，郁闭度估测值绝对误差依次为0.038、0.030、0.024、0.004。钩梢和郁闭度等对提取结果没有显著影响。  结论  基于多尺度分割的监督分类方法总体精度最高，估算绝对误差最小，能够满足快速、准确提取并估测毛竹林林分郁闭度的要求，且适用于不同经营类型的毛竹林。图2表6参28

Estimation method of Phyllostachys edulis forest canopy density based on UAV visible image

  Objective  Due to the growth and management characteristics of Phyllostachys edulis forest, the canopy density of Ph. edulis forest is a very important factor in the management. This study aims to explore the estimation method of canopy density of Ph. edulis forest based on unmanned aerial vehicle (UAV) visible image, which can achieve real-time and rapid acquisition of Ph. edulis forest canopy density.   Method  The visible light image of Ph. edulis forest of common rotor UAV was taken as the research object. 4 mature digital image processing methods were adopted, namely, threshold classification based on pixel, supervised classification based on pixel, threshold classification based on multi-scale segmentation, and supervised classification based on multi-scale segmentation. 36 sample plots with different truncation conditions and canopy density were selected. Using the existing software and MATLAB programming, the Ph. edulis canopy area in each sample plot was rapidly extracted, and then the canopy density was estimated. The estimation accuracy of the canopy density of each method was compared with the true value calculated by visual interpretation, and the performance of the 4 methods under different truncation and different canopy density conditions was compared and analyzed.   Result  The overall accuracy of threshold classification based on pixel, supervised classification based on pixel, threshold classification based on multi-scale segmentation, and supervised classification based on multi-scale segmentation was 91.81%, 92.96%, 93.47% and 98.86%, respectively. The absolute error of the estimated value of canopy density of the 4 methods was 0.038, 0.030, 0.024 and 0.004, respectively. Truncation condition and canopy density had no significant effect on the extraction results.   Conclusion  The supervised classification method based on multi-scale segmentation has the highest overall accuracy and the smallest absolute error. It can quickly and accurately extract and estimate the canopy density of Ph. edulis forest, and is suitable for different management types of Ph. edulis forest. Ch, 2 fig. 6 tab. 28 ref.]