Research On The Method Of Forest Volume Inversion Based On Multi Feature Extraction

Forest stock is an important index to measure the source of biological life materials supplied by forest,which can reflect the change and growth of forest resources,and is of great significance to the sustainable development and management of forest.The traditional methods of obtaining forest stock are mainly the first and second kind of forest resources investigation.The accuracy of the two methods is high,but there are great limitations in time and space,and the field investigation is needed.Artificial field work is not only time-consuming and laborious,but also requires the operators to master the relevant knowledge of forest resources investigation to a certain extent,which increases the time and difficulty of forest resources investigation of category ? and ? from other aspects.With the progress of science and technology,especially the development of remote sensing technology,the way of obtaining forest volume has changed from human investigation to inversion by using modern technology.The research on the estimation algorithm of forest stock can not only make people grasp the change of forest resources information timely and accurately,strengthen the efficient management of forest resources,but also provide the basis for the sustainable development of forest resources.However,due to the low accuracy of the feature factors extracted from the inversion model,the volume estimation results obtained by the inversion method are slightly inferior to the results of the first and second class forest resources investigation.In order to solve the problem of low accuracy of the volume estimation model,the old mountain construction area of Maoershan Forest Farm was selected as the research area,and 120 small groups of mixed coniferous and broad-leaved forest were selected as the research original data.The characteristic factors that can determine the volume of forest by the number of tree crowns and height of trees obtained from remote sensing image data were studied.On this basis,the partial least square method and principal component analysis regression method were used The correlation model can effectively improve the estimation accuracy of forest volume model.In order to improve the extraction accuracy of the characteristic factors of forest volume:(1)an improved watershed segmentation algorithm is proposed,which uses the classification method of maximum likelihood and support vector machine to divide the coniferous and broad-leaved mixed forest into coniferous forest aggregation area and broad-leaved forest aggregation area;then the image with higher classification accuracy is selected as the next segmented data image;finally,the maximum entropy threshold marking method is used to mark the data image respectively The optimal average thresholds of coniferous forest and broad-leaved forest are obtained,and the maximum entropy optimal average thresholds of coniferous forest and broad-leaved forest are compared.This method not only reduces the over segmentation caused by noise,but also reduces the over segmentation phenomenon caused by different crown sizes of coniferous and broad-leaved forests.(2)The elevation value is extracted by the canopy height model,and the average elevation value of the sample land with zero value removed is used to estimate the tree height and DBH.(3)This paper proposes an estimation method which is not limited to the traditional definition of canopy density.Combining with many factors including the number of tree crowns and terrain factors,the method of principal component analysis regression is used to establish the equation of canopy density.In this paper,while making full use of the aerial survey image,the details of the aerial survey image and the characteristics of various factors are kept to the maximum extent.The experimental results show that after the improved watershed segmentation algorithm,the crown segmentation accuracy is 80.03%,the tree height,DBH and canopy density estimation accuracy are 97.34%,91.27%and 83.18%respectively,which improves the extraction accuracy of feature parameters,and the estimation accuracy of the established forest volume estimation model can reach 88.43%.