Classifying Hyperspectral Images with Capsule Network and Active Learning

Deep neural networks have been widely applied to the task of hyperspectral image(HSI) classification, and most of them achieved high accuracy. However, the successes of these models rely heavily on the large number of labeled training samples, which are hard to obtain. To reduce the training cost of the HSI classifiers and improve the classification performance, we proposed an HSI classification method based on active learning (AL) and capsule network (CapsNet). We trained and validated our classifier with two HSI datasets and compared it with other advanced classifiers. By comparison, we found that our classifier could achieve higher accuracy with fewer training samples.