摘要
借助多源遥感影像数据进行森林资源分类是提高分类精度和实现森林资源精细识别的有效途径。以江西省吉水县为研究区,基于Landsat-8 OLI和GF-2遥感影像数据,在对采样数据进行光谱、纹理分析的基础上运用支持向量机分类法对森林资源进行分类,并对两种影像分类结果精度评价后采用替换法进行多源遥感影像分类。结果表明:(1)Landsat-8 OLI和GF-2影像森林资源分类最佳纹理窗口和方向分别为7×7、135°和19×19、135°。(2)GF-2影像加入纹理特征后阔叶林、针阔混交林的生产者精度和用户精度均高于Landsat-8 OLI影像,灌木林生产者精度与用户精度分别低于和高于Landsat-8 OLI影像,而竹林的生产者精度(63.33%)和用户精度(67.86%)均低于Landsat-8 OLI影像的生产者精度(90%)和用户精度(77.14%);GF-2影像加入纹理特征后总体分类精度从80.97%提高到87.04%,提高了6.07%,而Landsat-8 OLI影像只提高了2.23%。(3)采用替换法得到的多源遥感影像分类结果总体精度达到88.87%,比GF-2影像分类结果精度高1.83%。因此,综合分析多源遥感影像在森林资源分类中表现出的差异性并采用替换法对多源分类结果进行优势结合有助于提高遥感影像的森林资源分类精度。
It is feasible and effective way to improve the accuracy of forest resources classification and to realize the fine recognition of forest resources by means of multi-source remote sensing image data.Based on Landsat-8 OLI and GF-2 remote sensing image data,the method of support vector machine(SVM)was used to classify the forest resources in Jishui County,Jiangxi Province after spectrum and texture analysis of sampled data.Then,a substitution method was used to combine the two image’s classification results after evaluating their accuracy. The results were as follows.(1) The optimal texture window and direction for forest resources classification of Landsat-8 OLI and GF-2 images were 7×7,135° and 19×19,135°,respectively(.2)The producer accuracy(PA)and user accuracy(UA)of broadleaf forest and conifer-broadleaf forest classified by GF-2 image after added texture feature were higher than those of Landsat-8 OLI image.The PA and UA of shrubwood were lower and higher than those of Landsat-8 OLI image,respectively.The PA(63.33%)and UA(67.86%)of bambooforest classified by GF-2 image were lower than the PA(90%)and UA(77.14%)of Landsat-8 OLI image’s bamboo forest.The overall accuracy(OA)of GF-2 image increased from 80.97% to 87.04%,which increased by6.07%,while that of Landsat-8 OLI image increased by only 2.23%(.3)The OA of multi-source remote sensing image classification by substitution method was 88.87%,which was 1.83% higher than that of GF-2 image classification.So,comprehensive analysis of the differences in the multi-source remote sensing images in forest resources classification and the use of substitution method to combine the advantages of multi-source classification results will help to improve the forest resources classification accuracy of remote sensing images.
作者
徐辉
潘萍
杨武
欧阳勋志
宁金魁
邵锦锋
李琦
XU Hui;PAN Ping;YANG Wu;OUYANG Xun-zhi;NING Jin-kui;SHAO Jin-feng;LI Qi(College of Forestry,Jiangxi Agricultural University,Nanchang 330045,China;Foreign Fund Project Office of Jiangxi Forestry Department,Nanchang 330038,China)
出处
《江西农业大学学报》
CAS
CSCD
北大核心
2019年第4期751-760,共10页
Acta Agriculturae Universitatis Jiangxiensis
基金
国家自然科学基金项目(31360181,31760207)
亚洲开发银行CCF(气候变化基金)江西赠款项目(0229-PRC)~~
