基于主动学习的高光谱图像分类方法

Hyperspectral image classification based on active learning

高光谱图像监督分类中,为了避免休斯效应需要大量的训练样本,但在实际应用中对样本进行标注成本非常高,因此,得到高质量的训练样本显得十分重要。提出一种基于主动学习的高光谱图像分类方法,通过对区域关注度的统计,有效地结合图像光谱和空间特性,基于主动学习方法获取信息量较大的训练样本,从而较大幅度提高了分类的精确度。实验结果表明,所提算法比传统的随机取样监督分类法和主动学习方法在分类精确度上有较大的优势。

Most supervised classification methods require large training samples to avoid the well-known Hughes effect. However, labeling samples is often very expensive in actual world applications. In order to reduce the number of training samples, high-quality training samples are extremely important. A hyperspectral image classification based on active learning was proposed. It provided a new calculation method for concerning region attention degree to combine spectral and spatial characteristics of the image effectively, and used active learning method to obtain training set with the most abundant information and improved the classification accuracy ultimately. The experimental results show that the proposed method performs particularly well for the classification of hyperspectral images, when compared to random sampling supervised classification method and active learning approaches.