Efficient Deep Learning of Nonlocal Features for Hyperspectral Image Classification

Deep learning based methods, such as Convolution Neural Network (CNN), have demonstrated their efficiency in hyperspectral image (HSI) classification. These methods can automatically learn spectral-spatial discriminative features within local patches. However, for each pixel an HSI, it is not only related to its nearby pixels but also has connections far away from itself. Therefore, incorporate the long-range contextual information, a deep fully convolutional network (FCN) with efficient non-local module, named ENL-FCN, proposed HSI In framework, FCN considers entire input and extracts information receptive field. The module embedded unit capture information. Different traditional neural networks, extracted specially designed criss-cross path computation efficiency. Furthermore, by using recurrent operation, pixel's response aggregated all of HSI. benefits our ENL-FCN are threefold: 1) incorporated effectively, 2) be freely plug-and-play fashion, 3) much fewer parameters requires less computational resources. experiments conducted on three popular datasets demonstrate that method achieves state-of-the-art classification performance lower cost comparison several leading networks