基于地基LiDAR高度指标的小麦生物量监测研究

Monitoring of Wheat Biomass Based on Terrestrial-LiDAR Height Metric

为探讨Li DAR监测作物生物量的可行性和方法,以小麦为研究对象,通过田间试验获取关键生育期的小麦Li DAR点云高度指标和地上部生物量,基于幂函数回归与支持向量回归、利用十折交叉验证法分别进行特征选择和模型构建,选取各算法最优的全生育期小麦地上部生物量监测模型,并在测试集上对模型的预测能力进行检验与比较。结果表明:利用H95和生育期特征所构建的全生育期支持向量回归模型精度最高,训练集上决定系数R2达到0.814,测试集结果(R^2=0.821,RMSE为1.730 t/hm^2,RRMSE为32.77%)表明,模型具有较好的准确性;利用Hmean所构建的全生育期幂函数回归模型决定系数R2达到0.809,测试集结果(R^2=0.815,RMSE为1.760 t/hm^2,RRMSE为33.33%)也表明模型具有较好的准确性;高度指标估测小麦生物量具有先天局限性,所构建模型较适宜于监测小麦地上部生物量小于10 t/hm^2的情况,在超过10 t/hm^2的样本集上,95%的模型预测值被低估,RMSE呈指数增长;生育期特征有利于提升监测模型预测精度。

Rapid,nondestructive and accurate monitoring of crop biomass is of great significance for crop productivity estimation and intelligent management.In order to explore the feasibility of monitoring crop biomass with light detection and ranging(LiDAR),LiDAR point cloud height metrics and aboveground biomass were obtained from field trials at key growth stages of wheat.Then based on the power function regression and support vector regression,the ten-fold cross-validation method was used to pick features and construct models,and the optimal wheat aboveground biomass monitoring models for whole growth period were selected respectively.Finally,the prediction abilities of the two models were tested and compared on the test set.The results showed that the support vector regression model constructed by the H 95 and growth period provided the highest accuracy with an R^2 being as high as 0.814 on training set,and its test results were with R^2 of 0.821,RMSE of 1.730 t/hm^2,and RRMSE of 32.77%,which indicated that the model possessed good accuracy and adaptability.The power function regression model constructed by H mean provided an R 2 of 0.809,and its test results were with R 2 of 0.815,RMSE of 1.760 t/hm^2,and RRMSE of 33.33%,which also indicated that the model possessed good accuracy and adaptability.Estimation of wheat biomass by a height metric had inherent limitations,and the two models were more suitable for monitoring the aboveground biomass of wheat values less than 10 t/hm^2.On the whole sample set with aboveground biomass exceeding 10 t/hm^2,95%of the predicted values of the models were underestimated and RMSE was increased exponentially.The feature of growth period was helpful to improve the prediction accuracy of the monitoring model.