Research On Deep Learning Estimation Method Based On Forest Stand Volume Of Ground Images

Tree species classification and stock volume estimation in the forest stand are two very important tasks in forest management.At present,the main research method is to complete the ground survey by forestry personnel,but the ground survey work based on manpower is time-consuming and labor-intensive,and the efficiency is low.Therefore,in order to ensure the working efficiency within a short period of time,this paper proposes to use deep learning to classify the tree species in the ground forest image under complex background,and use the pixel information of the forest trees in the image to estimate the accumulation Measurement.In this paper,the SLR camera is used to shoot the forest stands vertically and the canopy images are taken.First,the U-Net model in the deep learning model is used to semantically segment the forest stand longitudinal images and identify the types and number of trees contained in the images And detection,and extract the pixel values of each tree from the segmentation results,as one of the factors for the subsequent accumulation modeling,and calculate the canopy degree from the crown image,as another factor for the accumulation modeling.Due to the time-consuming,labor-intensive and low efficiency of the existing methods for measuring forest stock volume,remote sensing methods on a large scale have low accuracy and cannot meet the needs of accurate forestry estimation.Therefore,this article will explore the feasibility of CNN neural network model for real-value regression of forestry data,and adopt random forest(RF),kernel extreme learning machine(KELM),support vector machine(SVM),least squares in machine learning Multiplicative support vector machine(LSSVM)and nonlinear mixed effect models are used to build up stock volume estimation models.The experimental results are compared and tested to prove the feasibility of the method in this paper.The research results show that the accuracy of the test set for semantic segmentation of U-Net forest image based on deep learning is 96.03%;the average Io U is 86%.Among the above accumulation volume estimation models,the estimation model based on the CNN model shows strong predictive power,and the model R~2 is 0.98.The training accuracy and prediction accuracy of the CNN model are higher than the other four neural network models mentioned above.Therefore,this paper concludes that the method based on ground forest image and deep learning can not only accurately classify trees,but also use neural network model to model and estimate the forest stock volume,which overcomes the traditional methods that are too dependent on certain Statistical assumptions and other issues can quickly realize the estimation of forest stock volume,and provide new ideas for the study of forest growth and harvest prediction.