Branch Information Extraction Of Apple Tree Based On LiDAR Data

The branches of apple trees are the basis for the growth of leaves,buds,flowers,fruits and other organs,and the appropriate number and length of branches are the guarantee of normal growth,flowering and fruiting of fruit trees.At the same time,apple tree branch information(length/number)is an important indicator of tree growth and yield.therefore,accurate acquisition of apple tree branch information is of great significance for orchard production.Based on Terrestrial Laser Scanning(TLS)and Backpack Light Detection and Ranging(Backpack-LiDAR)point cloud data of apple tree,this study analyzed the sensitivity of five key parameters in the Quantitative Structure Model(QSM),obtained the sensitive parameters for the reconstruction of apple trees with low height and complicated branches,and systematically described the subsequent parameter optimization methods.At the same time,based on the optimized QSM,the sample trees are modeled and extracted their branch information,and compared with the measured values.The main conclusions of this study include:(1)the principle of QSM was introduced,and the parameter sensitivity analysis was carried out.The results showed that the first-order sensitivity index and the total sensitivity index of parameter PD2Min were much more than 0.5,and it was the most sensitive to the extraction of apple tree branch information.In this study,the change of the number of parameter combinations set by sensitivity analysis has little effect on the change of sensitivity index,but the larger the number size was,the higher the reliability of sensitivity index was.(2)based on TLS point cloud data,the parameters were optimized and apple tree branch information was extracted,and the effect of point cloud density on modeling was discussed.The results show that when the parameter PD2Min=0.6cm was used,the relative error of branch information extraction was the smallest,and the best value of PD2Min is 0.6cm.Under the best condition of PD2Min,the relative errors of the first-order branch length and quantity extraction results were 7.43%and 12.00%respectively,the relative errors of the second-order branch length and quantity extraction results were 16.75%and 9.67%,and the relative errors of the third-order branch length and quantity extraction results were 35.84%and 23.81%respectively.The relative errors of the extraction results of total branch length and number were 15.34%and 2.89%,respectively.When the point cloud densits were 60063/m3,48050/m3,30031/m3,12638/m3,6268/m3,1658/m3,respectively,the relative errors of branch length extraction of the first,second and third-order were all about 10%,20%and 30%,the relative errors of the number of branches of the first,second and third-order were all about 10%,10%and 20%,and the relative errors of the information content of the total branches were all about 10%.When the point cloud space density was 807/m3,only the first branch information extracted was reliable.The accuracy of branch information extracted by TreeQSM increased with the increase of point cloud density in a certain range.For the TLS point cloud data in this study,when the point cloud spatial density was higher than 1658/m3,the extraction results of the first,second,third-order and total branch information are almost unchanged.(3)based on Backpack-LiDAR point cloud data,the sample trees were modeled and branch information were extracted.The results showed that in PD2Min=1.2 cm,the root mean square error(RMSE)and relative root mean square error(NRMSE)of branch information extraction were minimum,and PD2Min was the best.Compared with the measured value,there was obvious correlation between the extracted branch information and the measured value,and the correlation of the second and fourth-order branches were not as good as that of the first-order branches,while the third-order branch information and the total branch information were obviously underestimated.The R2 of the first-order branch length extraction result was 0.397,the RMSE was 3.43m,the NRMSE was 14.62%.For the first-order branch number,the R2 was 0.4351,the RMSE was 1.30,and the NRMSE was 11.96%.Based on the Backpack-LiDAR point cloud data,the first-order branches can be completely reconstructed and the information can be extracted.