本文選題：遙感　切入點(diǎn)：大區(qū)域　出處：《林業(yè)科學(xué)研究》2017年02期

【摘要】：[目的]利用遙感影像的時(shí)效性和宏觀性特點(diǎn),基于證據(jù)理論組合多分類規(guī)則的方法快速和高效地實(shí)現(xiàn)大區(qū)域植被遙感分類。[方法]首先,依據(jù)辨識(shí)框架的概念設(shè)計(jì)分類系統(tǒng),并采用大區(qū)域樣本快速采集方法提取訓(xùn)練樣點(diǎn);其次,將多個(gè)單分類規(guī)則得到的植被類型特征影像歸一化處理為基本概率賦值作為表達(dá)對(duì)各類型信任程度的證據(jù)源數(shù)據(jù),再將不同證據(jù)源的信任度信息依據(jù)證據(jù)理論組合;再次,將組合結(jié)果依據(jù)最大信任度原則確定植被類型;最后,在中國(guó)植被圖與中國(guó)土地覆蓋圖的類型一致區(qū)域隨機(jī)布點(diǎn)作為驗(yàn)證樣本。[結(jié)果]各單分類器分類結(jié)果的總體精度范圍為60%70%,兩兩規(guī)則組合分類結(jié)果的總體精度范圍為70%80%,3個(gè)規(guī)則組合分類結(jié)果的總體精度達(dá)到80.84%。[結(jié)論]組合多分類規(guī)則的證據(jù)理論分類方法可以提高分類精度;參與組合的單分類器精度越高,相關(guān)證據(jù)源越多,組合分類結(jié)果精度越高。
[Abstract]:[objective] based on the characteristics of timeliness and macroscopicity of remote sensing images, the classification system of vegetation in large area is designed quickly and efficiently based on evidence theory and multi-classification rules. [methods] first of all, according to the concept of identification framework, a classification system is designed. The training points are extracted by using the fast sampling method of large area samples. Secondly, the normalized vegetation type image obtained by multiple single classification rules is treated as the basic probability assignment as the evidence source data to express the degree of trust in each type. Then, the trust information of different evidence sources is combined according to the evidence theory; thirdly, the combination results are determined according to the principle of maximum trust. Finally, Random distribution of points in the area consistent with the type of vegetation map and land cover map in China is used as the verification sample. [results] the overall accuracy range of the classification results of each single classifier is 60 pand, and the total precision range of the results of the combination of pairwise rules is 60 p. The overall accuracy of the three rule combination classification results is 80.844.Conclusion the classification accuracy can be improved by combining the evidence theory classification method with multiple classification rules. The higher the accuracy of the combined single classifier, the more relevant evidence sources, the higher the accuracy of the combined classification results.
【作者單位】： 中國(guó)林業(yè)科學(xué)研究院資源信息研究所;
【基金】：國(guó)家863計(jì)劃課題(2012AA102001)
【分類號(hào)】：Q948;TP79
，

本文編號(hào)：1683189