Study On Sichuan Province Hilly Areas Woodland Remote Sensing Classification Methods Based On SPOT5

Using remote sensing survey of forest resources with a macro has comprehensive, short-period, repeatability and low cost characteristics. So the remote sensing technology in the investigation and monitoring of forest resources has a very important significance and role. However, in the remote sensing technology, there are still many technical issues needing to be addressed in the investigation and monitoring of forest resources, especially in the distribution of the high degree of fragmentation Sichuan province hilly areas. Carrying out the survey of forest resources in application of existing remote sensing image classification method, often results that the classification accuracy is not high and classification result of the application is not strong, This also needs a lot of artificial visual modifications result in classification and increases investigative workload. Therefore, the remote sensing classification methods and techniques of hilly woodland has become a serious problem for investigation and monitoring of forest resources in the region.This research took advantages of SPOT-5remote sensing data basing on the hilly region of Sichuan Nanbu County town of Lite as the study area and focused on several key issues in remote sensing classification, including the identification of the best band combination, remote sensing classification and maximum likelihood, support vector machines the accuracy evaluation and decision tree classification method. This got the following main results:(1) Getting the SPOT-5image band on the study area, the amount of information, standard deviation, and correlation characteristics such as value has statistical analysis. Through qualitative analysis and best-exponential factor value OIF calculated, the results showed that123bands of SPOT-5remote sensing images was one of the best3-band combinations with the greatest amount of information and the minimal information redundancy.(2) With the remote sensing processing software envi4.4support, Selecting the maximum likelihood classification, support vector machine classification and decision tree classification of remote sensing classification tests three different classification methods and gets classification results Figure. Finally, Using matrix and kappa coefficient gets accuracy verification and applicability evaluation with, the results of three methods of classification confusion. The overall classification accuracy of three classification methods followed by the maximum likelihood classification (80.83%)<support vector machine classifier (85%)<decision tree classifier (89.24%), the overall accuracy of the calculation of the three methods of classification are more than80%to satisfy the classification requirements;(3) The applicability analysis results showed that:The accuracy of maximum likelihood classification and support vector machine classification were lower, the decision tree classification got highest precision. Arbores, bamboo forest, open woodland, water, construction land and arable land successively reached96%,89%,82%,93%,98%and80%of classification accuracy.The analysis indicated that, the study maximum likelihood had lower classification accuracy,,the taxonomy support vector machine and decision tree method has higher one. The overall classification of the decision tree classification accuracy reached89.24%and kappa coefficient reached0.8581,which applicability accuracy reached above80%. This shows that the knowledge-based combination of normalized vegetation index and texture features a decision tree classification method is suitable for the hilly area of remote sensing of forest land classification.