论文部分内容阅读
多示例学习以示例组成的包作为训练样本,学习的目的是预测新包的类型。从分类角度上,处理问题的策略类似于以均质对象为基本处理单元的面向对象影像分类。针对两者之间理论和方法相似性,将多样性密度多示例学习算法与面向对象方法相结合用于高分辨率遥感图像分类。以图像分割方法获取均值对象作为示例,利用多样性密度算法对样本包进行学习获取最大多样性密度示例,最后根据相似性最大准则对单示例包或是经聚类算法得到的新包进行类别标记,以获取最终分类结果。通过与SVM分类器的比较,发现多样性密度算法的平均分类精度都在70%以上,最高可达96%左右,且对小样本问题学习能力更强,结果表明多示例学习在遥感图像分类中有着广泛应用前景。
Multi-Sample Learning Package consisting of examples as a training sample, the purpose of learning is to predict the type of new package. From a classification point of view, the strategy of dealing with problems is similar to the object-oriented image classification using homogenous objects as the basic processing unit. Aiming at the similarity of theory and method, the multi-sample density learning algorithm and object-oriented method are applied to high resolution remote sensing image classification. Taking the average object as the image segmentation method as an example, the sample density is calculated by using the density density algorithm. Finally, according to the maximum similarity criterion, a single sample bag or a new package obtained by the clustering algorithm is labeled For the final classification result. By comparing with SVM classifier, it is found that the average classification accuracy of diversity density algorithm is above 70% and up to 96%, and the learning ability of small sample problem is stronger. The results show that the multi-sample learning in remote sensing image classification Has a wide range of applications.