Tree Species And Age Groups Classification Based On GF-2 Image

As the study area in Jiangle state-owned forest farm of Fujian province,the important management trees of South Collective Forest Area:Cunninghamia lanceolata?Pinus massoniana and Phyllostachys edulis as the research object,this study aimed to explore the potential of China's high spatial resolution GF-2 Satellite in tree species and age groups classification.The research provided references for promoting the application of domestic GF-2 satellite images in forest resource investigation and management.The main results are as follows:(1)Based on the fact that there are few researches on preprocessing and tree species classification,the pretreatment process and fusion methods of GF-2 imagery were explored.The results showed that Gram-Schmidt and HCS fusion methods are most suitable for GF-2 which achieved the best results.(2)The canopy spectral curves of main tree species(age groups)were measured.The spectrum reflection of each trees were significantly different,height and slope of the "red edge"in spectral curve also showed differences.Combined with the GF-2 bands,it appeared that Green and NIR are the key bands to tree species and age groups classification.(3)The overall accuracy of the scheme combined with 4-direction texture attributes is up to 87.4%(Kappa=0.85),better than the scheme combined with all-direction texture attributes of 85.2%(Kappa=0.82);the overall accuracy of scheme using only spectral properties is 78.4%(Kappa=0.75),better than schemes that only texture attributes were used with overall accuracy less than 65%.Results showed that the combination of rich texture information improved the classification accuracy greatly.(4)Contrast experiments based on three classifiers of the maximum likelihood,support vector machine and random forest,the results showed that random forest achieved best effect on tree species classification under attributes screening,(overall accuracy=87.4%,Kappa=0.85);Support vector machine performed best in high-dimensional attribute features(overall accuracy=86.1%,Kappa=0.86);the maximum likelihood method leaded to general appearance with Hughes.(5)The growth morphology and canopy characteristics of tree species determined the effect of texture attributes on classification.Texture attributes have little effect on the accuracy of Pinus massoniana classification,but improved the accuracy of young-aged and middle-aged Cunninghamia lanceolata greatly.Mixed phenomenon was appeared in age groups of Cunninghamia lanceolate while young and mature forests were completely distinguished.It showed that GF-2 has great potential in tree species and age groups classificasion,which is an ideal data source for forest investigation in future.