| Forest,herein referred to one of the most important natural resources on the earth,provides the basis material for human survival and is recognized as the fundamental of sustainable development of forestry.The extraction of change information based on sub-compartments and the land types of forest is beneficial to effectively and accurately master the current situation and change information of the forest,provide a basis for the implementation of forest management plan,standardize and guide forest scientifically management for realizing the sustainable development of forest resources.This study used remote sensing images of Jiande,Zhejiang province in November 2013 and October 2014,and the forest inventory data in 2007 and the supplementary survey data in 2013 of Jiande to compare methods for extracting change information of forest sub-compartments based on the statistical distribution of vegetation index and the Zone model.Then,this study implemented fine extraction of forestland types based on pixel,Object-oriented and prior knowledge and used confusion matrix to evaluate the methods synthetically to explore a suitable method for fine extraction of forestland types,and support investigating the changes of forest resources and updating data of sub-compartments.The results showed that:(1)The extraction method of change information based on the Zone model performed better than the one based on the statistical distribution of vegetation index.The positive detection rate of the Zone model was 80.29%,false detection rate was 83.45%,and missed detection rate was 19.42%.Compared with the method based on the statistical distribution of vegetation index,the positive detection rate increased by 8.64%,the false detection rate decreased by 26.26%,and the missed detection rate decreased by 8.64%.The results showed that the method based on the Zone model was useful and greatly improved the efficiency of investigation on the change of forest resources and the update of sub-compartments.(2)The method that using change information of forest sub-compartments and historical survey data as prior knowledge to extract the forestland types performed better than the method based on pixel and object-oriented.The former overall accuracy was 90.23% and Kappa was 0.86%.From the visual effect of fine extraction results and the efficiency of extraction process,the results of the methods based on pixel were elaborate,but there were many errors and omissions.The large noise,and inaccurate classification of boundary pixels or mixed pixels resulted a large workload after classification.The results of the methods based on object-oriented had intact information,but there were omissions in scattered land types.The method based on prior knowledge combined pixel-based and object-oriented methods,so the results of which based on prior knowledge could not only ensure that the information in the fragments was not lost,but also ensure the integrity of the patches.(3)The dependence of the method of fine extraction based on prior knowledge on training samples was low and the result was reliable.Using system sampling technique to verify area of the main forest land types of the result based on prior knowledge,the accuracy of arbor forest was 99.98%,that of national special shrub forest was 92.64%,and the accuracy of general shrub forest was 89.62%.Moreover,this paper combined extraction of changing information and forestland types with forest sub-compartments to get the change information and forest land types of forest sub-compartments,which is practical significance to investigate the change of county-level forest land resources and update the data of sub-compartments,and more accorded with the actual demand of our country.Therefore,the fine extraction of forestland types based on prior knowledge had good value in practical production and application. |
PDF Full Text Request