摘要
BP algorithm of neural net is used more in remote sensing data classification. One of drawbacks of BP algorithm is the overall low function when the net is training. To avoid this kind of problem, the paper introduces the tolerant rough set for classification-preprocessing the training data to reduce the influence elements of the training convergence in order to improve the net training successful rate. ETM+ data of Beijing in May 2003 is selected in the study. ETM+ data before and after classification preprocessing, respectively, are used for BP (Back propaga-tion) training. The result shows that such a preprocessing not only compensates the drawback of BP algorithm when processing ETM+ data but also improves classification accuracy.
BP algorithm of neural net is used more in remote sensing data classification. One of drawbacks of BP algorithm is the overall low function when the net is training. To avoid this kind of problem, the paper introduces the tolerant rough set for classification-preprocessing the training data to reduce the influence elements of the training convergence in order to improve the net training successful rate. ETM+ data of Beijing in May 2003 is selected in the study. ETM+ data before and after classification preprocessing, respectively, are used for BP (Back propaga-tion) training. The result shows that such a preprocessing not only compensates the drawback of BP algorithm when processing ETM+ data but also improves classification accuracy.
基金
This work was supported by the National Natural Science Foundation of China(Grant No.4037086)
China“863”Project(Grant No.2003AA135080).
