Active deep densely connected convolutional network for hyperspectral image classification

Abstract

Deep learning based methods have seen a massive rise in popularity for\nhyperspectral image classification over the past few years. However, the\nsuccess of deep learning is attributed greatly to numerous labeled samples. It\nis still very challenging to use only a few labeled samples to train deep\nlearning models to reach a high classification accuracy. An active\ndeep-learning framework trained by an end-to-end manner is, therefore, proposed\nby this paper in order to minimize the hyperspectral image classification\ncosts. First, a deep densely connected convolutional network is considered for\nhyperspectral image classification. Different from the traditional active\nlearning methods, an additional network is added to the designed deep densely\nconnected convolutional network to predict the loss of input samples. Then, the\nadditional network could be used to suggest unlabeled samples that the deep\ndensely connected convolutional network is more likely to produce a wrong\nlabel. Note that the additional network uses the intermediate features of the\ndeep densely connected convolutional network as input. Therefore, the proposed\nmethod is an end-to-end framework. Subsequently, a few of the selected samples\nare labelled manually and added to the training samples. The deep densely\nconnected convolutional network is therefore trained using the new training\nset. Finally, the steps above are repeated to train the whole framework\niteratively. Extensive experiments illustrates that the method proposed could\nreach a high accuracy in classification after selecting just a few samples.\n