Research On Classification Of Tree Species Based On GF-2

Forests are the most important terrestrial ecosystems.It is important to master tree species composition and distribution in forests for studying and utilizing forest ecosystems.In recent years,the successive launches of the Gaofen(GF)series and the Ziyuan(ZY)series satellites have greatly enriched the data sources of high spatial resolution remote sensing images in China.In order to further promote the application of Chinese high spatial resolution data in the classification of forest tree species and explore the impact of feature selection and classification methods on tree species classification results at the same time,we used the main part of Badaling National Forest Park in Yanqing District of Beijing as the research area and used the six GF-2 images as the data source.Based on the hierarchical classification,the related forest tree classification research was carried out.The main research contents are as follows:(1)In order to explore the efficient and accurate tree species classification method in multitemporal data,this study first carried out tree species classification research by superimposing 6 scene images.This research used support vector machine-recursive feature elimination,C5.0 algorithm and feature space optimization to accomplish feature optimization at first,and accomplished the object-oriented classification of arbor species from four feature dimensions using support vector machines(SVM),C5.0 decision tree(C5.0DT)and random forest(RF)three classifiers.Finally,this research achieved good results with overall accuracy between 73.50% and 89.00% and the Kappa coefficient between 0.70 and 0.87.The results showed the C5.0 feature optimization method took the shortest time(0.67s)and features selected by it could be applied to the highest classification accuracy(89.00%);the C5.0DT performed the worst in all feature dimensions,the overall accuracy of SVM is 1.37% higher than the RF in average,and overall performance of the SVM is more stable,but the RF is more efficient;three classifiers were both insensitive to feature dimension,but good feature optimization results will still have a large impact on the classification efficiency(Highest improvement was 95.29%)and classification accuracy(Highest improvement was 9.50%)of classifiers.(2)At the same time,in order to explore the effects of different combining images,classification features and classifier classification results,three kinds of single temporal data and four kinds of multi-temporal data were constructed by three images of six.According to the C5.0 feature optimization as well as two classifiers including SVM and RF,we finished the objectoriented classification of eight arbor species of different temporal and feature dimensions respectively and finally achieved good results with overall accuracy between 63.5% and 83.5% and the Kappa coefficient between 0.57 and 0.81.The results showed that the choice of temporal stage would affect the classification results.The results based on multi-temporal were generally better than that on single temporal stage.There were obvious precision differences between different image combinations of multi-temporal as well as different single temporal stage.It is notable that feature optimization played a positive role in the improvement of classification accuracy.SVM was relatively stable across different temporal and feature dimensions,which should be given priority when in single temporal and classification features are difficult to distinguish different tree species directly,while it should be noted that SVM is prone to overfit.RF is not prone to significant overfit,but it was more dependent on the quality of classification features and would get better results under favorable image combination.