Deep Pyramidal Residual Attention Networks For Hyperspectral Image Classification

In recent years,the classification of hyperspectral and high spatial resolution images is becoming more and more popular.Hyperspectral images and high spatial resolution not only contain a lot of semantic information of spatial context,but also bring a lot of spectral information.In addition,due to the high-dimensional characteristics of hyperspectral image data cube and its complex spatial structure,the difficulty of feature extraction of hyperspectral image is increased.Convolutional neural networks(CNNs)have been commonly used for hyperspectral image(HSI)classification,often exhibit good performance in image processing tasks.However,at the edge of each land-cover area,an HSI cube often contains several pixels whose land-cover labels are different from that of the center pixel.These pixels are called interference pixels,which will weaken the resolution of spectral spatial characteristics as well as decrease classification accuracy.In order to mitigate these issues,a deep pyramidal residual attention network(PRAN)is specially designed for the HSI data.Our new model pursues to improve the spectral-spatial features uncovered by the convolutional filters of the network.Specifically,the proposed residual-based approach gradually increases the feature map dimension at all convolutional layers,grouped in pyramidal bottleneck residual blocks,in order to involve more locations as the network depth increases while balancing the workload among all units,preserving the time complexity per layer.Therefore,the diversity of high-level spectral-spatial attributes can be gradually increased across layers to enhance the performance of the proposed network with the HSI data.Moreover,attention mechanism is added to the residual module,which makes the effective feature map with heavyweight and the invalid or small effect feature map with small weight can achieve better results.In order to evaluate the effectiveness of the model,we conducted experiments on four famous HSI datasets.Experiments on several public HSI databases show that the performance of the proposed PRAN is better than the existing algorithms.