Study On Subtropical Forest Monitoring Method Based On Uav Remote Sensing And AI Algorithm

Forest resources monitoring is an important basic work to effectively grasp the status quo and dynamic changes of forest resources and ecological environment.UAV(Unmanned Aerial Vehicle)has the characteristics of flexibility,efficiency and convenience.It overcomes the limitations of traditional remote sensing technology in time and space resolution and can meet the demand of dynamic observation and customized remote sensing.According to the current forest resources monitoring system,it is necessary to promote the upgrading of forest resources monitoring methods,promote the application of artificial intelligence in forestry,and aim at the monitoring needs of the stem volume and biodiversity change caused by the expansion of moso bamboo in the evergreen broad-leaved forest.Based on the data obtained by UAV lidar and digital camera,this study uses artificial intelligence algorithm to explore the extraction and estimation methods of forest resource information.From single tree scale monitoring methods for moso bamboo expansion to the stand scale stem volume and biodiversity estimates the research on evergreen broad-leaved forest,for UAV-Li DAR remote sensing combined with artificial intelligence method to provide reference for application in forest resources monitoring,and also to the UAV-Li DAR technology in the application of evergreen broadleaf for the research field.The main research contents and results of this paper are as follows:(1)Point cloud data were obtained based on lidar,and the applicability of CHM data-marked watershed algorithm and PCS segmentation algorithm based on normalized point cloud data in single tree segmentation in evergreen broad-leaved forest and moso bamboo forest was compared.The results showed that both the PCS algorithm and the marker watershed algorithm based on the appropriate CHM resolution could identify the number of moso bamboo trees effectively,and the F1-score was close to0.75.In evergreen broad-leaved forest,the recognition accuracy of PCS algorithm is better than that of watershed algorithm,but the overall accuracy is still low.The ratio of tree height to canopy radius is a good indicator to eliminate the oversegmentation of PCS algorithm in broad-leaved forest.At the same time,based on the filed measurement of tree height,it is found that the tree height obtained by lidar is close to the field measurement value on the basis of accurate spatial position identification.(2)Single tree scale type of vegetation canopy photo sample set was made by using single tree segmentation result combined with 0.1 m resolution digital orthogonal map,compared the use of multiple state of art deep convolutional neural networks,with transfer learning which loaded weight from Image Net to make classification of vegetation type,overall accuracy of Dense Net121 reached90%,evergreen broad-leaved tree,deciduous broad-leaved tree,bamboo got an accurate over 90%,conifer is limited by texture resolution F1-score is 0.72.The research shows that the combination of single tree information extracted by lidar with RGB texture information and the identification of vegetation type by deep convolutional neural network is an effective application of UAV combined with artificial intelligence in forestry.It can effectively monitor the spatial distribution of broad-leaved forest and moso bamboo,and regular monitoring can master the extent of moso bamboo expansion.(3)in the land of 6 hectares of evergreen broad-leaved forest based on the investigation,at 20 m x20 meters sample scale,a total of 150 samples was generated,the correlation between vertical features witch extracted from high density Li DAR point cloud with stem volume was analysed,whicn found that intensity variables is strong negative related to the stem volume.After the significance(P<0.05)was used to clean part of the characteristic features,the random forest machine learning algorithm was used to obtain the accuracy of R~2=(0.61-0.84)and the error of RMSE%=(17%-28%).Through the analysis of variable importance,the study found that the accumulative intensity variables and forest gap rate is important to the precision prediction of random forest algorithm,at the same time,based on the part of dependence analysis and individual conditional expectation analysis found that based on high density models for predicting the point cloud's volume of random forests build time,important control variable is forest gap rate,sampling should consider the data distribution of forest gap rate.(4)For biodiversity monitoring,a shifting window sampling method was introduced to resample200 m×300 m rectangular plots at different scale of 400m~2,1600 m~2 and 3600 m~2 with a window size of 1×1,2×2,3×3 and a moving step of 20m×20m.150(10×15),126(9×14)and 104(8×13)quadrates were generated,respectively.At the same time,four groups with the diameter threshold dbh greater than or equal to 1cm,5cm,10cm and from 1cm to10cm were calculated on the Richness,Shannon and Pielou indexes,and the application of the Scaling diversity index and Shannon index on the 3600 m~2sample scale were compared.The interpretation ability of lidar in the biodiversity index of evergreen broad-leaved forest was discussed.The results showed that the optimal diversity estimation was obtained at the sample plot scale of 60 m×60 m.For a single index,Pielou uniformity index has the strongest explanatory power,and the estimation accuracy of validation set(R~2=0.842)is obtained based on PLSR model for dbh?1cm.For Richness,the interpretation ability of the lidar with dbh?10cm was the strongest with R~2=0.809.According to Shannon index,it was found that the accuracy of estimation of 1cm?dbh?10cm(R~2=0.785)and dbh?10cm(R~2=0.876)was higher for the overall estimation.At the same time,the Scaling diversity index including spatial scale parameters was introduced to express the abundance of richness more effectively compared with Shannon index,and the estimation accuracy based on UAV-LIDAR was higher than that of Shannon index.In general,lidar has great application potential in ecological monitoring of evergreen broad-leaved forest.UAV lidar combined with camera can effectively monitor bamboo forest expansion.The height and intensity matrix generated by lidar are important variables for the modeling of stem volume and biodiversity estimation,and accurate estimation results are obtained based on artificial intelligence algorithm.Using UAV lidar for regular monitoring can effectively provide an important basis for the assessment of ecological civilization construction objectives.