欢迎访问林业科学,今天是

林业科学 ›› 2016, Vol. 52 ›› Issue (6): 54-65.doi: 10.11707/j.1001-7488.20160607

• 论文与研究报告 • 上一篇    下一篇

多源数据林地类型的精细分类方法

任冲1, 鞠洪波1, 张怀清1, 黄建文1, 郑应选2   

  1. 1. 中国林业科学研究院资源信息研究所 北京 100091;
    2. 甘肃省小陇山林业调查规划院 天水 741020
  • 收稿日期:2015-12-21 修回日期:2016-02-18 出版日期:2016-06-25 发布日期:2016-07-04
  • 基金资助:
    国家高技术研究发展计划(863计划)"数字化森林资源监测技术"项目(2012AA102001)。

Multi-Source Data for Forest Land Type Precise Classification

Ren Chong1, Ju Hongbo1, Zhang Huaiqing1, Huang Jianwen1, Zheng Yingxuan2   

  1. 1. Research Institute of Forest Resource Information Techniques, CAF Beijing 100091;
    2. Xiaolongshan Forest Inventory and Planning Institute of Gansu Province Tianshui 741020
  • Received:2015-12-21 Revised:2016-02-18 Online:2016-06-25 Published:2016-07-04
  • Contact: 鞠洪波

摘要: [目的] 探讨复杂中山区域、多源数据支持下,高空间分辨率遥感影像林地类型层次化精细分类方法,以促进高分辨率遥感数据在森林资源调查与监测方面的深入应用。[方法] 以嘉陵江上游甘肃省小陇山林业实验局百花林场为研究区,以SPOT5和高分一号(GF-1)遥感影像为主要数据源,综合利用影像光谱特征、植被指数特征、纹理特征与时相特征、地形特征、森林资源"二类调查"成果数据与林相图等辅助信息,及典型地类与主要森林类型外业调查样本数据,发展针对暖温带典型天然次生林区、复杂山区地形条件下高空间分辨率遥感影像林地类型多层次信息提取与森林类型精细识别的有效方法。在分析不同时相影像光谱特征的基础上,构建并优选归一化植被指数(NDVI)、比值植被指数(RVI)、比值短波红外指数(RSI)、差值植被指数(DVI)4种植被指数特征和均值(ME)、同质性(HOM)、非相似性(DIS)、信息熵(ENT)、角二阶距(ASM)、相对峰值(RK)6种纹理特征,引入与主要森林类型空间分布相关的DEM高程值、坡度、坡向3个敏感地形因子,利用不同林地类型时相动态特征和辅助信息特征,在不同层次影像上分别采用适于该层待分信息类别的阈值法、支持向量机(SVM)、多分类器组合(MCC)、人工神经网络(ANN)分类方法,将各层分类结果合并获得整个研究区林地类型精细分类图。最后,采用分层随机抽样的独立检验样本对分类结果中7类林地类型进行精度验证,并对5类主要森林类型精细识别结果进行面积统计,与"二类调查"及影像解译结果各类型面积统计值进行对比分析,进一步从整体上检验分类方法的有效性和分类结果的可信度。[结果] 本文所发展的分类方法对林地类型信息提取精度较高,有林地、其他林地、苗圃地等7类林地类型总体分类精度达92.28%,总Kappa系数为0.8996;油松林、华山松林、日本落叶松林、栎类落叶阔叶林、其他落叶阔叶混交林5类主要森林类型面积统计结果的平均相对精度为92.4%。[结论] 多源数据支持下的多层次林地类型精细分类方法是一种有效的林地类型信息精准监测方法,具有精度高和可信度高的优势,且森林类型精细识别详细程度达到优势树种(组)级别,是解决复杂山区林地类型精细分类与森林类型精细识别的一种有效手段,可满足森林资源调查、变化监测、数字更新等林业应用需求。

关键词: 多源数据, 精细分类, 多层次信息提取, 多分类器组合, 林地类型

Abstract: [Objective] This paper presents a new forest land type precise classification method based on hierachical classification strategy using high spatial resolution remote sensing image and multi-source auxiliary data in complex mountainous area, which locates in Baihua Forest Farm, Xiaolongshan Forestry Experimental Bureau, Gansu Province.[Method] In the test area, SPOT5 HRG2 and two scenes of GF-1 PMS2 images, ancillary information such as forest resources inventory results, forest form map and forest distribution map, and field measurement sample point data of land cover types and fine forest types were applied in classification process. The multivariate characteristics based on high resolution remote sensing images, including image spectral features, vegetation indices, textural features and image time-phase changed features, and topography characteristics, were used as the significant indicators to develop multi-level information extraction methods and forest type fine identification methods, which were particularly suitable for the complex mountainous terrain area of typical natural secondary forest region in warm temperate zone. Four vegetation indices,including NDVI(normalized difference vegetation index),RVI(ratio vegetation index),RSI(ratio shortwave infrared index)and DVI(difference vegetation index)and six textural features,that is ME(mean),HOM(homogeneity),DIS(dissimilarity),ENT(entropy),ASM(angular second moment)and RK(relative kurtosis)were developed and selected. Three topography characteristic factors related to spatial distribution of major forest types such as DEM elevation slope and aspect were also extracted and used in the hierarchical classification process. Threshold method, support vector machine(SVM), multiple classifier combination(MCC) and artificial neural network(ANN) classification methods were developed and applied in different levels of image. Classification results of different layers were combined into forest types fine classification result map of the whole research area. Finally, independent test samples of seven forest types based on stratified random sampling were selected and the confusion matrix and Kappa coefficient were examined to evaluate the accuracy of classification results. In order to further evaluate the validity of classification method and reliability of classification results on the whole, the performance of the classification method was discussed by comparing the statistic area about five main forest types based on classification results with the statistic results of forest resources inventory and image visual interpretation.[Result] The results showed that the method proposed in this paper had a good performance in forest type information extraction. Overall accuracy of seven forest types, including closed forest land, other types of forest land, nursery land and so on reached 92.28% and overall Kappa coefficient was 0.8996. Average relative accuracy of area statistic results of five main forest types of fine identification, including Pinus tabulaeformis, Pinus armandii, Larix kaempferi, Oak species dominant broad-leaved deciduous forest and other (hardwood species dominant) deciduous broad-leaved mixed forest, reached 92.4%.[Conclusion] The results of this study suggest that the hierarchical information extraction method based on multi-source data support is an effective approach of precise classification of forest land types, fine identification of forest types and accurate monitoring of forest resources, especially in mountainous area of complicated topography conditions. The proposed method in this paper have advantages in fine identification of forest land types with high accuracy and high reliability, and the detail degree of fine identification reaches dominant tree species, which could fully meet the needs of forestry applications such as forest resources investigation, forest land change monitoring and thematic map digital update, etc.

Key words: multi-source data, precise classification, hierarchical information extraction, multiple classifier combination, forest land type

中图分类号: