基于距离相关系数和KNN回归模型的森林蓄积量估测研究

【目的】探究Landsat8 OLI数据和KNN算法在森林蓄积量估测中的潜力。【方法】以湖南省湘潭县为研究区,采用Landsat8 OLI数据和同时期的二类调查数据,通过距离相关系数筛选特征,分别采用线性回归模型(MLR)、K-近邻模型(KNN)、距离加权KNN模型(DW-KNN)和优化欧式KNN模型(FW-KNN)对森林蓄积量进行估测。使用十折交叉方法进行精度检验,对检验结果进行对比分析。【结果】3种KNN模型的估测结果均高于传统的线性模型,并且在3种KNN模型中,FW-KNN算法效果最好,决定系数达到0.69,为3种模型中最高;3种KNN模型中,本研究优化欧氏距离KNN模型的估测精度最高,其均方根误差为30.3%,相比于传统KNN模型的均方根误差降低了5.1%,相比于DW-KNN模型降低了3.3%。【结论】采用DW-KNN蓄积量估测结果明显优于其他两种模型,说明通过特征与蓄积量的相关性优化样本间的距离是一种可行的KNN优化方法。

Forestry volume estimation based on distance coefficient and KNN regression model

【Objective】Explore the potential of Landsat 8 OLI data and KNN algorithm in forest stock estimation.【Method】Taking Xiangtan county of Hunan province as the research area,the Landsat8 OLI data and the inventory data of forest resources in the simultaneous period were used to screen the features by distance correlation coefficient,using linear regression model(MLR),K-nearest neighbor model(KNN)and distance weighted KNN model.(DW-KNN)and the optimizing euclidean distance KNN model(FW-KNN)estimate forest stocks.The ten-fold crossover method was used for the accuracy test,and the test results were compared and analyzed.【Result】The estimation results of the three KNN models are higher than the traditional linear model,and among the three KNN models,the FW-KNN algorithm works best,and the coefficient of determination reaches 0.69,which is the highest among the three models.Among the three KNN models,The improved KNN model has the highest estimation accuracy,and its root mean square error is 30.3%.Compared with the traditional KNN model,the root mean square error is reduced by 5.1%,which is 3.3%lower than the DW-KNN model.【Conclusion】The estimation of DW-KNN accumulation is better than the other two models,indicating that it is a feasible KNN improvement method to optimize the distance between samples by the correlation between features and accumulation.