中文摘要：

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。

英文摘要：

In multiple instance learning,the bags are used as training samples,and the goal of learning is predict the label of new bags.The idea of multiple instance learning is quite similar to the object-oriented image classification,which takes homogeneous object as basic processing unit,so it is feasible to combine multiple instance learning with object-oriented way to classify the high resolution remote sensing image.In this paper,Diverse Density（DD） algorithm is used to classify the high resolution remote sensing image according to the object oriented image classification paradigm.Homogeneous objects are generated by image segmentation method first,and then objects used as instances,get the maximum diverse density instance by training bags with DD,so the label of new bags which single instance considered as a bag or obtained by clustering method can be determined by the distance similarity criterion.Compared with the advanced SVM classifier,the classification approach consisting of diverse density algorithm and object oriented can get higher classification accuracy,average accuracy is higher than 70%,the highest one is 96%,and its capability to small sample learning problem is also well.The result shows that multi instance learning based remote sensing image classification has a wide prospect.