Study On The Recognition Of Forest Resources Based On Landsat-8 OLI And GF-2 Data

With the development of remote sensing satellite,the recognition of large-scale forest resources by remote sensing technology has become an important technical means of forest resources investigation and monitoring.Taking Jishui County,Jiangxi Province as the research area and based on Landsat-8 OLI image and GF-2 image,the spectral and texture feature of land types in the area was compared by sampled data.Then,in order to analyse the effect of different methods and texture feature on the recognition of land types in remote sensing image,using the method of MLC(Maximum Likelihood Classification),SVM(Support Vector Machine),ANN(Artificial Neural Networks)extracted the two image's forest resources information and the substitution method was used to produce the multi-source recognition results.The main results and conclusions were as follows:(1)According to the main types of land use and the resolution of the image,the land types recognition system was established.The forest resources based on Landsat-8 OLI was divided into 6 types: coniferous forest,broad-leaved forest,coniferous and broadleaved mixed forest,bamboo forest,shrub forest and other woodland;the forest resources based on GF-2 was divided into 9 types: china fir,pine,broad-leaved forest,coniferous and broadleaved mixed forest,bamboo forest,shrub forest,cut-over area,immature forest,other woodland;the two images classified the non-forest resources into four types: cultivated land,water body,construction land and others.(2)Different selection of texture parameters(such as step size,window size and direction)has different effects on recognition results.By setting different texture parameters and using MLC to classify the two image's experimental area found that when the texture parameter(the step size,window size and the direction)of Landsat-8 OLI image and GF-2 image was 1,7×7,135° and 1,19×19,135° respectively,the OA(overall accuracy)was the highest.(3)The improvement of the OA of GF-2 image by texture features was significantly higher than that of Landsat-8 OLI image and the OA of SVM method was the highest.The OA of three methods(MLC,SVM,ANN)of Landsat-8 OLI image based on spectral feature was 82.59%,84.61% and 81.38% respectively.After adding texture feature,the OA was improved by 1.62%,2.23% and 0.6% respectively;the OA of three methods(MLC,SVM,ANN)of GF-2 image based on spectral feature was 77.13%,80.97% and 78.34% respectively.After adding texture feature,the OA was improved by 5.87%,6.07% and 6.89% respectively.(4)The improvement of the PA(producer's accuracy)and UA(user's accuracy)of GF-2 image forest resources types by texture features was significantly higher than that of Landsat-8 OLI image.After adding texture feature of Landsat-8 OLI image and using SVM method,the PA and UA of the coniferous forest and broad-leaved forest was improved by 3.27%,2.85% and 0.15%,0.83% respectively.But for the coniferous and broad-leaved mixed forest,bamboo forest,other woodland and shrub,only the UA was improved by 4.55%,2.14%,3.76%,3.25% respectively.The PA and UA of GF-2 image of china fir,pine,broad-leaved forest,bamboo forest and the coniferous and broad-leaved mixed forest after adding texture feature was improved by 8.25%,12.82%,11.43%,6.66%,5.88% and 9.66%,6.46%,7.44%,7.15%,18.85% respectively.But only the UA of immature forest and other woodland was improved by 4.48%,5.49% respectively and both of the two accuracy of shrub and cut-over area still remain unchanged.(5)Comprehensive analysis of the recognition differences of multi-source remote sensing images in forest resources recognition and the use of substitution method to combine the advantages of multi-source recognition results was helpful to improve the recognition accuracy of remote sensing images.With the help of spectral advantage,the bamboo forest's PA of Landsat-8 OLI image can reach 90%,and the UA was more than 77%,which was obviously higher than that of GF-2 image(63.33% and 67.86%).But with the help of texture advantage,the OA of GF-2 image was improved from 80.97% to 87.04%,increased by 6.07%,which was 3.84% higher than that of Landsat-8 OLI image after adding texture feature.By using the substitution method combined with the advantages of the two kinds of image recognition,the OA of multi-source recognition was 88.87%,increased by 1.83%.