一种利用结构特征的高分辨率遥感影像种植园自动提取方法

A Structure Feature for Automatic Extraction of Plantation from High-resolution Remote Sensing Imagery

卫星遥感在土地监测中发挥了重要作用,而数据处理技术应具有良好的适用性和较高的精度水平以满足应用需求。针对目前种植园提取算法适用性较差、数据依赖性高、自动化程度低、算法复杂及特征冗余等问题,本文提出了一种新型种植园结构特征,多尺度多方向结构指数,通过对其进行阈值化并结合形态学处理实现了种植园的自动提取。对多种植模式、背景植被化程度和结构复杂度各不相同的全色和多光谱种植园影像分别设计了3组试验,结果证明,与传统方法相比,本文方法适用性更强,对多方向、多尺度、基元变形、种植线变形、种植模式规律性差、背景高度植被化等诸多复杂的种植园等均能进行有效提取,整体精度达到90%以上。此外,试验结果中,全色影像与多光谱影像精度相当,表明本文方法具有较低的光谱依赖性,因此在数据选择和应用上具有更大的灵活性。

Satellite remote sensing is an invaluable tool to manage land resources. However,data analysis procedures should satisfy the good adaptability,wide application prospects and high accuracy levels demanded by users. This study presented a novel multi-scale and multi-direction structure index（ MMI） to describe the structure feature of plantation caused by cultivation. Plantation are extracted by performing a threshold on the MMI feature map,and combined with morphological operators to refine the extraction results. We designed three groups of experiments to test our method,each group used panchromatic and multispectral imagery respectively with various cultivation mode,different vegetated background and structure complexity. The results show our method is much more adaptive on plantation extraction than traditional methods. It is efficient for various complex plantations,e. g.multi-direction,multi-scale,highly vegetated backgrounds,low regularity of planting mode with deformation of textons and planting lines,the accuracy results exceed 90%. And panchromatic images achieve accuracies as high as multispectral images,which indicate our method has low dependence on spectrum,thus it is more flexible for data selection and application.