基于PCA变换和神经元网络分类方法的中国森林制图研究

MAPPING OF NATURAL FOREST IN CHINA BASED ON PRINCIPAL COMPONENT TRANSFORM AND NEURAL NETWORK CLASSIFICATION

主要讨论基于NOAA AVHRR数据生成的NDVI时间序列数据与其他来源的地理数据结合进行中国森林细分类和制图的研究。首先利用ISODATA聚类方法对由NDVI时间序列数据派生的NDVI矩阵变量因子进行土地覆盖类型的分类,然后利用已有的植被类型图、TM影像以及地面样点数据作为参考依据,进行类合并,获得中国森林边界。然后用PCA方法对NDVI时间序列数据进行信息增强与压缩处理,以排除各种干扰因素,提高分类精度。最后结合部分地理数据和地面样点调查数据,利用神经元网络方法进行中国森林分类,并依据种属和物候学特征、中国气候区划图以及国家植被分类二、三级分类系统,进行重新归类,得到最后的1 km中国森林分类图。分类结果表明,所用方法能够更加细致地划分森林类型,并且能保留一个相对较高的分类精度。

This paper presented the research on natural forest classification and mapping in China based on NDVI time series datasets derived from NOAA AVHHRR pathfinder data, and some ancillary geographical data. The authors first had a land cover classification by applying an ISODATA clustering method on some NDVI matrices indices from NDVI time series data, then detected the natural forest border through classes combination based on vegetation type map, TM image and field sampling point data. Secondly, in order to improve the classification accuracy, the authors performed a PCA transform on the NDVI time series data to remove noises. Finally, combined with other geographic data and field investigation data, the authors trained and constructed a neural network classifier to get the original forest classification map in China, then based on forest types, phenology character, China climatic division map and the national second and third classification system on vegetation, through classes combing, the forest type map in China was obtained. The final classification result showed that the classification method used in the paper can not only give a more detailed forest classification, but also have a better classification accuracy.