Abstract
In this paper, we design a simple, robust and powerful neural network architecture for hyperspectral image (HSI) classification, where state-of-the-art results can be achieved with only a small number of training samples. The proposed framework is a relation network (RN), whose objective is to learn the similarity between pairs of samples (pixels) in the same hyperspectral images. Once trained, the proposed relation network is able to classify each testing sample in the hyperspectral images by computing the relation scores between the testing and training samples. Evaluations on three real HSI data sets show that the relation network outperforms the previous state-of-the-art deep learning models.
Bin Deng
Lecturer
My research interests include hyperspectral image processing, pattern recognition and machine learning.