Research On Forest Resources Information Estimation Model Based On Deep Learning

Forests are the largest ecosystem on Earth's land and account for about one-third of the Earth's land area.Forest biomass for about 90%of the total biomass of the Earth's terrestrial ecosystem and plays a decisive role in the global carbon cycle,which can be seen as one of the most valuable resources on Earth.Therefore,it is of great significance and value to study how to estimate the forest resource information quickly and efficiently on biomass and stock volume.Remote sensing(RS)technology as a non-contact detection technology,can achieve rapid,large-scale,periodic detection of forests,is very suitable for forest resource information estimation research.In this paper,the remote sensing image of "high GF-1 satellite is used as the data source,and the Dayekou River in Heihe of Gansu Province is the research area.The forest resource information in the study area is estimated.The main contents include:1,Analyze the correlation between the characteristics extracted from the remote sensing image and the forest biomass.The spectral information and vegetation index were extracted from the GF-1 remote sensing image as the feature variables of the estimation model.The correlation between the feature variables and the forest biomass was analyzed.The results showed that the correlation is very weak between feature extracted from the remote sensing images and the forest biomass,so when using remote sensing technology for forest resource information estimation,the non-linear method is more appropriate.2,The Deep Learning(DL)method was introduced into the field of forest resource information estimation,and a method of forest resource information estimation based on deep learning was proposed.As a nonlinear learning method,deep learning has achieved good results in many fields.Based on the study of the theory of deep learning,this paper proposes a method to construct the forest biomass estimation model by using the deep learning method.The experimental results show that this method is feasible and its estimation accuracy is superior to the traditional forest biomass estimation method based on BP neural network.3,Through analyzing the experimental results,the problems and solutions of the estimation model were discussed when the forest resource information is estimated by using a small number of training samples.The measurement of forest structural parameters is very complex,leading to the fact that the number of training samples in the forest resource information estimation experiment is usually less.When using non-parametric models such as neural networks,the quality of training samples will have a great impact on the estimation results.In view of this situation,this paper presents a method of randomly dividing train samples and test samples,and carrying out multiple experiments to verify the model.When estimating the biomass for the entire study area,the neural network parameters with the highest accuracy can be used to estimate the best accuracy.