遥感信息
遙感信息
요감신식
2015年
2期
43-49
,共7页
城镇植被群%分层分类%多描述符空间%植被分类
城鎮植被群%分層分類%多描述符空間%植被分類
성진식피군%분층분류%다묘술부공간%식피분류
city vegetation population%hierarchical classification%multi-descriptor space%vegetation classification
针对城镇绿化植被受可用土地限制，具有分布零散、结构复杂、植被群类型繁多等特点，以及建筑物和其他设施的阴影遮挡等加剧了植被群分类的困难，经典分类方法常常难以适应需求的困境，该文提出了以分层分割／分类和多描述符空间分类对城镇绿化植被群分类的方法，即采用一次采样，多层分类，利用分类要素间的层次关系，从粗到精依次分类，以降低分类难度和提高精度。通过对这些图斑做层之间关系的逻辑分析，以精度较高的边界取代较低者，可以得到精度较好的目标类图斑。此种方法可用于改善草地、乔灌木、阔叶植被、针叶植被、落叶植被、常绿植被等植被群之间的可分性。仿真测试表明，该算法对于高分辨遥感图像城镇绿化植被群分类具有一定的普适性，划分植被群的总体精度高于85％，比经典单层分类方法提高了5％～10％。
針對城鎮綠化植被受可用土地限製，具有分佈零散、結構複雜、植被群類型繁多等特點，以及建築物和其他設施的陰影遮擋等加劇瞭植被群分類的睏難，經典分類方法常常難以適應需求的睏境，該文提齣瞭以分層分割／分類和多描述符空間分類對城鎮綠化植被群分類的方法，即採用一次採樣，多層分類，利用分類要素間的層次關繫，從粗到精依次分類，以降低分類難度和提高精度。通過對這些圖斑做層之間關繫的邏輯分析，以精度較高的邊界取代較低者，可以得到精度較好的目標類圖斑。此種方法可用于改善草地、喬灌木、闊葉植被、針葉植被、落葉植被、常綠植被等植被群之間的可分性。倣真測試錶明，該算法對于高分辨遙感圖像城鎮綠化植被群分類具有一定的普適性，劃分植被群的總體精度高于85％，比經典單層分類方法提高瞭5％～10％。
침대성진녹화식피수가용토지한제，구유분포령산、결구복잡、식피군류형번다등특점，이급건축물화기타설시적음영차당등가극료식피군분류적곤난，경전분류방법상상난이괄응수구적곤경，해문제출료이분층분할／분류화다묘술부공간분류대성진녹화식피군분류적방법，즉채용일차채양，다층분류，이용분류요소간적층차관계，종조도정의차분류，이강저분류난도화제고정도。통과대저사도반주층지간관계적라집분석，이정도교고적변계취대교저자，가이득도정도교호적목표류도반。차충방법가용우개선초지、교관목、활협식피、침협식피、락협식피、상록식피등식피군지간적가분성。방진측시표명，해산법대우고분변요감도상성진녹화식피군분류구유일정적보괄성，화분식피군적총체정도고우85％，비경전단층분류방법제고료5％～10％。
Constrained by limited urban land resources and the needs in artificial aesthetics,urban landscape vegetation is often characterized by scattered distribution,complex structure,various species and the shaded scene resulting in the severe challenges to the classification of vegetation population from remotely sensed imagery.To solve this problem,this paper presents an algorithm to classify vegetation populations such as the lawn,trees,broadleaf,conifers,deciduous and evergreen plants by using hierarchical classification (HC)in multi-descriptor space,that is,conducting multi-layer classifications in accordance with the hierarchical relationship of the classes and using different combinations of mathematical descriptors for the classifications. For example,the two classes of with and with no vegetation cover will firstly be separated.This will be followed by the classification between grass and trees (and/or shrubs)from the previous vegetation-covered class.After this the separation between broadleaf and conifers plants or between deciduous and evergreen plants from the above tree/shrub classes can be reached.By using the inheritance relationship between these classes in different layers,the accuracy of classification can further be improved because the boundary of a patch of a class in a certain layer can be replaced with a more accurate one of the same class in another layer.The whole algorithm was tested by MATLAB simulation.It is revealed that the overall accuracy (OA)is about 85% by using HC and has 5% to 10% better than that by using conventional single-layer classification.
