基于机器学习的辽东山区落叶松林蓄积量估算研究

Estimation of Larix Larix Forest Stock Based on Machine Learning in Mountainous Areas of Eastern Liaoning Province

作为森林资源信息的采集工具,遥感技术能及时有效的获取准确信息,是了解森林资源现状和获取森林结构参数的理想手段。以三类设计的作业小班为基础,小班中心为质点做25 m×25 m样方,取其公顷蓄积量作为因变量,提取影像各波段像元亮度值(digital number, DN)、植被指数、地形因子作为自变量指标,利用多元线性回归算法和人工神经网络建立了模型估算森林蓄积量。并利用实地调查的蓄积数据反演精度,对其算法的优劣进行了评价。经检验得知:多元线性回归平均绝对误差为35.68,相对均方根误差为39.32%;神经网络平均绝对误差为30.45,相对均方根误差为34.87%。典型地区精度可以达到80%。由评价表明:该模型使用良好,且误差在可接受范围内,利用人工神经网络方法获得的模型更优。

As a forest resource information collection tool,remote sensing technology can obtain accurate information timely and effectively,and it is an ideal means to get the status quo of forest resources and obtain forest structure parameters.In this paper,based on the small class of three-class design,the center of the class was used as the particle to make 25 m×25 m quadrats;the stock volume of hectares was extracted as the dependent variable,and the Digital Number value,vegetation index and terrain factor of each band of the image were extracted as the independent variable index.Multiple linear regression algorithm and artificial neural network were used to build a model to estimate forest stock.Based on the inversion accuracy of the storage data in field investigation,the advantages and disadvantages of the algorithm are evaluated.The mean absolute error of multiple linear regression is 35.68 and the relative root mean square error is 39.32%.The average absolute error of the neural network is 30.45,and the relative root mean square error is 34.87%.Typical area accuracy can reach 80%.According to the evaluation,the model is well used and the error is within the acceptable range.The model obtained by artificial neural network method is better.