Classifications of Multisource Remote Sensing Images

多源遙測影像之分類

范國泰(K. T. Fan) ； 曾裕強(Y. C. Tzeng)

《聯大學報》 7卷2期 (2010/12) Pp. 279-298

https://doi.org/10.29847/JNUU.201012.0013

摘要

同時使用由多個不同感測器所取得之遙測影像，預期可改善土地覆蓋分類的準確度。資料融合可利用多源資料間互補資訊的優點，而得到比僅用單一資料源更高的整體準確度，故尋求不同資料源間的互補資訊，是資料融合的一項非常重要的工作。本論文將採用多源分類器，以充分利用不同資料源間的互補資訊。但由於各資料源之單位不必然相同，因此當融合多源資料時，必須面臨資料尺度的問題。本論文採用兩種可克服資料尺度的問題的多源分類器（類神經網路及多分類器系統），並將其應用於多源遙測影像之分類，以展示並比較其分類效能。就多分類器系統而言，更以三種不同權重策略（均方根距離、平均距離、適應性閾值）來比較其分類效能。實驗結果顯示，類神經網路及多分類器系統均可大幅改善分類準確度，而多分類器系統的表現又優於類神經網路。此外，在多分類器系統中，適應性閾值權重策略則優於距離權重策略，且平均距離距離權重策略又略優於均方根距離權重策略。

關鍵字

資料融合 ； 類神經網路 ； 多分類器系統

並列摘要

The use of remote sensing images from various sensors is supposed to be able to improve land cover classification accuracies. The important prospect of data fusion is focused on exploiting the complementary information among different sensors. Data fusion can take advantage of the use of complementary information to obtain a better overall accuracy than using single data source only. In this paper, multisouce classifiers are adopted to fully utilize the complementary information among different data sources. Because the multisource data are not necessarily in common units and therefore scaling problems may arise at fusing different sources of data. To overcome the scaling problems, two types of the multisouce classifier, neural networks and multiple classifiers systems, are introduced. The performances of utilizing the multisouce classifiers to the application of multisource remote sensing images classification are demonstrated and compared. For the multiple classifiers systems, three different weighting policies (rms distance weighting, average distance weighting, and adaptive thesholding) are examined. Experimental results show that both the neural networks and multiple classifiers systems approaches can dramatically improve the classification accuracy. In addition, the multiple classifiers systems approaches outperforms the neural networks approach. Besides, the classification performance of the adaptive thresholding multiple classifiers system is better than those of the distance weighted multiple classifiers systems. Moreover, the classification performance of the average distance weighting multiple classifiers system is slightly better than that of the rms distance weighting multiple classifiers system.

並列關鍵字

data fusion ； neural networks ； multiple classifiers system

參考文獻

Benediktsson J. A., Swain, P. H., Ersoy, O. K.(1990).Neural network approaches versus statistical methods in classification of multisource remote sensing data.IEEE Transaction on Geoscience and Remote Sensing.28,540-552.

Benediktsson, J. A.,Chanussot, J.,Fauvel, M.(2007).Multiple classifiers in remote sensing: from basics to recent developments.Proceedings of the 7th International Workshop on Multiple Classifier Systems.(Proceedings of the 7th International Workshop on Multiple Classifier Systems).

Benediktsson, J. A.,Sveinsson, J. R.,Ersoy, O. K.,Swain, P. H.(1997).Parallel Consensual Neural Networks.IEEE Transaction on Neural Networks.8,54-64.

Bierman, G. J.(1977).Factorization Methods for Discrete Sequential Estimation.New York:Academic.

Breiman, L.(1996).Bagging Predictors.Machine Learning.24,123-140.