Research On Classification Of Fast-Growing Eucalyptus Based On TM By Using Remote Sensing Technology

In order to alleviate the contradiction between supply and demand in domestic timber market. Our country attaches great importance to create bases of fast-growing forest. Eucalyptus is one of the three admitted fast growing trees in the world, and it has widely cultivated in a lot of China’s southern provinces because of good strains, fast growth, the advantages of high efficiency and strong adaptation. The planting area of fast-growing eucalyptus in Guangxi ranks first in China, and it means an important strategic position in the forestry development of the whole nation. So it’s important to pay special attention to scientific management for fast-growing eucalyptus, which not only is the focus of forestry development in Guangxi, but also is the key to the forestry construction in our country. In this context, the paper took Jiepai Branch of Gaofeng Forestry Farm in Guangxi as study area, discussing classification technologies about eucalyptus based on TM, which would find the optimal classification way to classify fast-growing eucalyptus from forest and providing the basis for related department to make decision about management of fast-growing eucalyptus by using remote sensing.In the paper,7original bands from TM were added up and made curve of spectrum to analyze, which mastered separability ability of each TM band fully about main types as fast-growing eucalyptus, etc; On this basis, vegetation indexes, PCA, K-T Transform and MNF were performed by using TM in the study area,14characteristic bands were generated. With the addition of original6bands came from TM image, a total of20bands could be as data foundation of discussion about classification optimal band combination of fast-growing eucalyptus, then combined with typical spectral figure of features and classification evaluation of fast-growing eucalyptus came from OIF indexes to determine the optimal band combination; Ensured6class types (Water, Bare area, Eucalyptus, Cedarwood, Young stand, Economic forest) though considering the field research purposes and the reality in the study area, selecting the samples, and evaluating both purification and separability of them; Conducted classification of Fast-growing eucalyptus by means of4kinds of classifications (General supervised classification, BPNN, Decision Tree, LSMM), then tested and evaluated accuracy of outcomes.The main conclusions of this study:①Band4had low correlation with Band5in TM image, as a result they had a higher capacity of identifying main features in forest and could be the key bands to use in the classification of eucalyptus;②less characteristic of separability between eucalyptus age groups, so the classification test would concern on eucalyptus above young age instead of every eucalyptus age group;③The optimal classification combination of fast-growing eucalyptus were PC3\BI\GVI\TM3\TM4\TM5\DVI;④SVM was the best way to identify eucalyptus among all the general unsupervised classification methods, Kappa value was0.6755, producer’s accuracy of eucalyptus was80.42%; BPNN’s Kappa and producer’s accuracy were0.7074,74.30%separately; Quest Decision Tree with terrain factors did well versus other Decision Tree and the Kappa and producer’s accuracy were0.7069,86.00%; LSMM (linear decomposition of mixed pixels method) overall RMS error was0.0021, RMS error of eucalyptus was0.0059;⑤The relative error of SVM, BPNN, Quest Decision Tree of Ten bands and LSMM about fast-growing eucalyptus classification area were ranked in descending order as BPNN(28.72%)> SVM(13.57%)> LSMM(10.18%)> Quest Decision Tree of Ten bands(5.23%), the terrain factors involved in the Quest Decision Tree classification had the best effect on eucalyptus classification in the researched region.