Forest type identification by combining multi-temporal GF-6 WFV and Sentinel-2 data
[Objective]Because of the broken terrain and frequent cloudy and rainy weather,it is difficult to finely identify forest types in southern China.Exploring joint multi-source and multi-temporal remote sensing data is important for identifying forest types.[Method]This study took Xinfeng County of Jiangxi Province as the study area.Based on the Forest Resource Inventory of the Xinfeng County in 2019,eight forest types were identified,including pine forest,Cunninghamia lanceolata forest,broad-leaved forest,coniferous mixed forest,coniferous and broad-leaved mixed forest,bamboo forest,shrub forest and other forestry land.The random forest algorithm was used to compare the forest type identification ability of GF-6 WFV and Sentinel-2 in the same band(purple/dark blue,blue,green,red,near infrared,red edge)and different bands(yellow edge,short-wave infrared),and a combined spectral feature dataset was built.By combining the multi-temporal vegetation index feature dataset which was built by GF-6 WFV and Sentinel-2,the combined spectral feature dataset,texture features,and terrain features,a feature variable selection dataset was constructed using random forest and recursive elimination method for forest type identification.The accuracy of the identification results was verified by using confusion matrix and the actual distribution of forest types.[Result](1)The overall accuracy of the GF-6 WFV for the combination of blue,green and red band was 58.31%.With the addition of the purple,nearinfrared,red edge,yellow edge of GF-6 WFV band and short-wave infrared of Sentinel-2 band,the overall accuracy increased by 1.99%,8.90%,10.71%,1.50%and 14.10%,respectively.The overall accuracy of the blue,green and red band combination of Sentinel-2 was 54.68%.With the addition of the deep blue,nearinfrared,red edge,short-wave infrared of Sentinel-2 band and yellow edge of GF-6 WFV band,the overall accuracy increased by 3.30%,10.82%,12.92%,17.31%and 3.97%,respectively.(2)The overall accuracy and Kappa coefficient of feature variable selection dataset were 80.80%and 75.56%.The order of contribution degree was GF-6 WFV multi-temporal vegetation index,followed by sentinel-2 multi-temporal vegetation index,GF-6 WFV spectral feature,Sentinel-2 spectral feature,topographic feature and texture feature.The contribution rates were 40.44%,23.23%,18.12%,10.21%,4.61%and 3.39%,respectively.(3)The producer's accuracy of pine forest,Cunninghamia lanceolata forest,broad-leaved forest,coniferous mixed forest,coniferous and broad-leaved mixed forest,bamboo forest,shrub forest and other forestry land were 86.97%,85.60%,88.61%,9.43%,19.01%,53.60%,86.90%and 82.56%,respectively,and the user's accuracy was 81.42%,79.79%,77.57%,71.43%,81.82%,67.00%,87.74%and 82.88%,respectively.The identification results are relatively consistent with the actual forest type distribution in the study area.[Conclusion]The combination of multi-temporal GF-6 WFV and Sentinel-2 can integrate the advantages of multi-temporal and multi-source images and effectively improve the identification accuracy of forest types.
国家科技期刊平台
NSTL
万方数据
