改進(jìn)的多維遙感數(shù)據(jù)的自適應(yīng)遺傳超平面分類器算法
Self-adapted Genetic Hyperplane Classifier Algorithm for Multi-dimensional Remote Sensing Image
-
摘要: 在遙感圖像數(shù)據(jù)監(jiān)督分類方法中，普遍存在著通過(guò)大訓(xùn)練數(shù)據(jù)量提高分類精度的問(wèn)題。該文在筆者已經(jīng)實(shí)現(xiàn)的遺傳超平面方法基礎(chǔ)上，做了進(jìn)一步的改進(jìn)，這就使得這種遺傳超平面分類器可以使用了少量的訓(xùn)練數(shù)據(jù)進(jìn)行訓(xùn)練，而得到的分類精度與大訓(xùn)練數(shù)據(jù)量相比具有可以接受的差別；改進(jìn)了分類方法中使用主成分分析后再用兩個(gè)主成分進(jìn)行分類的做法，使用的原始數(shù)據(jù)為多個(gè)(3個(gè)以上)波段直接進(jìn)行分類，不但增加了分類輸入的信息量，而且簡(jiǎn)化了技術(shù)流程。同時(shí)，在不增加分類時(shí)間的情況下擴(kuò)展了算法分類的類別數(shù)。文中使用C/C++從底層實(shí)現(xiàn)了整個(gè)訓(xùn)練、分類、測(cè)試過(guò)程，通過(guò)對(duì)北京的ETM+數(shù)據(jù)進(jìn)行的分類實(shí)驗(yàn)及其分析表明該算法分類效果很好，完全可以達(dá)到實(shí)用的要求。
-
關(guān)鍵詞:
- 遙感圖像數(shù)據(jù);遺傳算法;超平面分類;分類精度
Abstract: There exists a problem that is using big quantity of training data to improve classification accuracy in remote sensing supervised classification methods. In this paper, advanced improvements are proposed for the implemented genetic hyperplane algorithm to get the advantage of using smaller quantity of training data and almost the same training effect. Generally, the principle component analysis is used to acquire the 2 principle components and the result is used to classify the data. Now that the improvement is that several bands (above 3) of remote sensing data are used simultaneously for the classification. Henceforth, the information quantity that input the classifier is incremental and the technological flow is simplified. At the same time, the number of classes from the algorithm is extended, while the time consuming is not incremental. The C/C++ is used to implement the whole process, which involve training, classification and test. The ETM+ data of Beijing is given for the classification and the good performance is acquired. The result shows that it can be fully used in practical. -
陳國(guó)良. 遺傳算法及其應(yīng)用. 北京: 人民郵電出版社, 1996, 第1章 第3章.[2]莊家禮, 陳良富. 用遺傳算法反演連續(xù)植被的組分溫度. 遙感學(xué)報(bào), 2001, 5(1): 1-7.[3]Davidson J L.[J].Xia Hua, Ashlock D. A comparison of genetic algorithm, gression, and Newtons method for parameter estimation of texture models.In Proceedings IEEE Southwest Symposium on Image Analysis and Interpretation, San Antonio, TX.1996,:-[4]李啟青, 馬建文, 哈斯巴干等. 遙感數(shù)據(jù)的遺傳超平面分割算法. 遙感學(xué)報(bào), 2003, 7(6): 485-489.[5]Pal S K, Bandyopahayay S, Murthy C A. Genetic classifiers for remotely sensed imges: comparison with standard methods. International Journal of Remote Sensing, 2001, 22(13): 2545-2569[6]Engebretson C, Davidson J, Ashlock D. Genetic Algorithms and Metropolis Algorithm for Model Selection.Technical Report, Departmeth of Electrical and Computer Engineering, Iowa State University, Ames, IA, 1996[7]Jianwen Ma, HuaDong Guo, Changlin Wang, Ge Chen, Vern Singhroy. Extraction of polymetallic mineralization information from multispectral thematic mapper data using the Gram-Schmidt Orthogonal Projection(GSOP) method[J].Interna- tional Journal of Remote Sensing.2001, 22(17):3323-3337[8]朱述龍, 張占睦. 遙感圖像獲取與分析. 北京: 科學(xué)出版社, 2000,第7章.[9]哈斯巴干, 馬建文, 李啟青等. 基于小波融合的ASTER數(shù)據(jù)自組織特征映射神經(jīng)網(wǎng)絡(luò)分類研究. 中國(guó)科學(xué)(D), 2003, 33(9): 895-902.[10]Cover T M, Hart P E. Nearest neighbour pattern classification. IEEE Trans. on Information Theory, 1968, 13: 21-27.[11]Schriever J R, Congalton R G. Evaluation seasonal variability as an aid to cover-type mapping from landsat thematic mapper data in the northeast. Photogrammetric Engineering and Remote Sensing, 1995(3): 321-327.[12]馬建文, 趙忠明, 布和. 遙感數(shù)據(jù)模型與處理方法. 北京: 中國(guó)科學(xué)技術(shù)出版社, 2001, 第1章.[13]李啟青, 馬建文, 哈斯巴干等. 基于貝葉斯網(wǎng)絡(luò)模型的遙感數(shù)據(jù)處理技術(shù)[J].電子與信息學(xué)報(bào).2003, 25(10):1321-瀏覽 -
計(jì)量
- 文章訪問(wèn)數(shù): 2312
- HTML全文瀏覽量: 145
- PDF下載量: 733
- 被引次數(shù): 0
下載:
