第4讲遥感图像解译之定量遥感方法.ppt

  1. 1、本文档共79页,可阅读全部内容。
  2. 2、有哪些信誉好的足球投注网站(book118)网站文档一经付费(服务费),不意味着购买了该文档的版权,仅供个人/单位学习、研究之用,不得用于商业用途,未经授权,严禁复制、发行、汇编、翻译或者网络传播等,侵权必究。
  3. 3、本站所有内容均由合作方或网友上传,本站不对文档的完整性、权威性及其观点立场正确性做任何保证或承诺!文档内容仅供研究参考,付费前请自行鉴别。如您付费,意味着您自己接受本站规则且自行承担风险,本站不退款、不进行额外附加服务;查看《如何避免下载的几个坑》。如果您已付费下载过本站文档,您可以点击 这里二次下载
  4. 4、如文档侵犯商业秘密、侵犯著作权、侵犯人身权等,请点击“版权申诉”(推荐),也可以打举报电话:400-050-0827(电话支持时间:9:00-18:30)。
查看更多
第4讲遥感图像解译之定量遥感方法

* Keeling curve Metric of climate variation e.g., glaciers, tree rings: different time and space scales global, seasonal to interannual to decadal. * * Dynamic range, atmospheric and soil background effects. * * Common approach to modeling plant growth – biologically consistent! * Piecewise continuous * * Supervised: Training Data * ~2000 Sites derived from hi-res imagery spanning all major regions ecosystems, but sampling based on “opportunistic” criteria * Training Sites—STEP Database (Muchoney et al., 1999; PERS) STEP: System for Terrestrial Ecosystem Parameterization Interpreted from Landsat ancillary data Key STEP Parameters Life form, cover fraction, leaf type, phenology, elevation, moisture regime, disturbance Simple description of site and type A confidence site near Pinsk, Belarus (20 x 20 km) * Basic Algorithm 1. Initialize w(i)t=1/N 2. At each iteration: 1. ?t = ∑ w(i) for incorrect predictions 2. w(i)t+1 = wt(i) ?t / (1– ?t) 3. Re-estimate tree 4. Weight for each tree B = ?t / (1– ?t) Where w(i)t = weight for the ith case in iteration t, and N is the total number of cases Optimizing Classification: Boosting (McIver and Friedl, IEEE TGARS 2001) Estimate multiple trees At each iteration, re-weight sample to focus on difficult cases Final classification Accuracy weighted vote across multiple trees * Post-Classification Processing (McIver and Friedl 2002, RSE) Application of Prior Probabilities Global priors to remove training site class distribution biases Moving-window priors from earlier products Use of external maps of prior probabilities to resolve confusions Agriculture/natural vegetation confusion in some regions Use of city lights DMSP data to enhance urban class accuracy Filling of Cloud-Covered Pixels from Earlier Maps Use of previous year product when there are not sufficient values to classify a pixel with confidence Supervised Classification: Technical Challenges Algorithms cannot compensate for inadequate features Use of spatia

文档评论(0)

dajuhyy + 关注
实名认证
内容提供者

该用户很懒,什么也没介绍

1亿VIP精品文档

相关文档