论文部分内容阅读
精选示例特征嵌入多示例学习(MILES)算法在对噪声较强的训练样本进行学习时表现出良好的性能,但其判断规则可能带来遥感影像分类结果的不确定性。针对这一问题,提出用Bagging和AdaBoost集成MILES的多示例集成学习算法,使用粗包细分、多样性密度和最大似然分类相结合抑制分类不确定性的方法,实现了高分辨率遥感影像分类中多示例学习与集成学习的组合。采用Quick Bird、IKONOS等高分辨率遥感影像进行试验,结果表明多示例集成学习能有效控制遥感影像分类结果的不确定性,具有良好的应用前景。
Featured Example Features The Embedded Multi-Instance Learning (MILES) algorithm shows good performance when learning noise-intensive training samples, but its decision rules may lead to uncertainties in the classification results of remote sensing images. Aiming at this problem, a multi-sample integrated learning algorithm based on MILES integrated with Bagging and AdaBoost is proposed. The method of suppressing the classification uncertainty by using the combination of rough packet segmentation, diversity density and maximum likelihood classification is proposed to realize high resolution remote sensing image The combination of multi-instance learning and integrated learning in the category. Experiments using Quick Bird, IKONOS and other high resolution remote sensing images show that the multi-sample integrated learning can effectively control the uncertainty of remote sensing image classification results and has good application prospects.