混合多分类器结合算法在遥感影像分类中的应用研究

Application research of hybrid multi-classifier combination algorithm on remote sensing image classification

为了提高遥感影像分类精度,从抽象级和测量级的两个层次出发,提出混合多分类器结合算法。该算法利用不同子分类器的分类结果及对各类别的分类精度,设定单个类别精度的阈值,选择最优子分类器,得到部分类别的最终分类结果;然后使用基于抽象级Bagging算法和测量级上的最大置信度进行多分类器结合。该算法应用于北京1号遥感影像的分类研究,结果表明该算法的总体精度和单个类别的分类精度比选用的子分类器都有明显的提高,是一种新的有效算法。

To enhance the accuracy of image classification, proposed hybrid multiple classifier combination method from abstract level and measurement level. Firstly, used different sub-classifiers and the accuracy of different classes. Secondly, after set the threshold value of different single class accuracy, obtained the optimal sub-classifier to get final results of partial classes. Thirdly, combined Bagging algorithm at abstract level and the most large confidence algorithm at measurement level to classify other classes. Finally, used this method in Beijing-1 image classification shows a better enhancement, and results also indicate that the hybrid multi-classifier combination algorithm is a newly effective algorithm.