Estimation Of Pinus Massoniana Forest Stock Volume Based On Multi-spectral UAV Image

Forest stock volume is not only an important index of forest resource asset evaluation,but also one of the important parameters for the accurate improvement of forest quality.However,in the current forestry production,the survey method of forest stock volume is still the traditional sample plot survey,which consumes a lot of manpower and material resources,it is urgent to improve the efficiency and precision of forest stock accounting.Multi-spectral UAV remote sensing has the characteristics of high efficiency and low cost.If it can be applied to improve the accuracy and efficiency of forest stock estimation,it will save a lot of manpower and material resources.This study took Pine massoniana plantation in Changshun Forest Farm,Changshun County,Guizhou Province as the research object.Ground surveys were conducted from July to November 2020.In October,the multispectral version of DJI Phantom 4 was used to fly the 41hm~2 study area,and the aerial photos were used for reconstruction to generate DOM.The local maximum method is used to extract the single wood position points.The 400 texture features of 10 window sizes and 20vegetation indices were extracted from DOM,random forest algorithm was used to filter the extracted factors.Combined with the location and volume of 1209 Pinus massoniana trees surveyed on the ground,linear model,gradient boosting regression tree,support vector machine,and random forest were used to predict the volume of single wood.The best model was selected according to the model’s performance.The forest volume in the study area was inverted by the single tree position and the best model.After summarized and analyzed,this study had the following conclusions:(1)Filtered by random forest algorithm,among 400 texture factors and 20vegetation indices,Ten factors including NIRMea59,GCor59,NIRDis15,NIRHom17,GMea79,REHom17,GSec17,NIRCor59,GHom15 and NIRSec15 have higher characteristic importance.The 10 factors were all texture factors.It can be seen that the texture factor has better explanatory for individual timber volume than the vegetation index.In addition,it can be seen from the selected factors that the texture factor of the NIR band accounts for 50%,the Green band accounts for 40%,the Red Edge band accounts for 10%,and the NIR band accounts for the largest proportion,indicated that the texture information of the NIR band had higher relationship with individual timber volume than vegetation index.(2)After model performance evaluation,among the four models,the random forest model had a mean square error MSE of 0.0052,a coefficient of determination R~2 of0.8846,MSE is lower than the other three models,and R~2 is higher than the other three models.It can be seen that compared with the Linear model,the Support Vector Machine model and the Gradient Boosting Regression model,the Random Forest model can better estimate the Pinus massoniana single timber volume.(3)According to the results of extracted individual plant positions of Pinus massoniana by the local maximum method,the NIR band had the highest accuracy,and the best filtering windows for the three density levels of Pinus massoniana were 5×5,7×7,and 13×13,respectively.The minimum extraction accuracy of individual plant positions was 70%,the maximum extraction accuracy was 91%,and the average extraction accuracy reaches 80%,indicated that the local maximum method could be used to extract the position and density of individual trees in Masson pine plantations.(4)The random forest model was used to invert the forest accumulation of Pinus massoniana in the study area.It was found that the forest accumulation of 9344 trees in the study area was a total of 3,952.77 m~3,and the ratio of the predicted value to the measured value of average individual plant volume and average accumulation per mu reached 97.6%and 89.70%.It shows that the estimation accuracy of multi-spectral UAV in the Pinus massoniana forest stock volume is relatively high,and it were feasible under the condition of ensuring the accuracy of two-dimensional reconstruction.In summary,multi-spectral UAV remote sensing technology was used to estimate Pinus massoniana plantation stock with high accuracy,which proved the feasibility of multi-spectral UAV remote sensing to achieve Pinus massoniana forest stock inversion.The research results can provide technical reference for remote sensing monitoring of Pinus massoniana forest stock by drone,and provide theoretical and technical guidance for the development of related research on multi-spectral UAV inversion of forest stock.