Research On Spatial-spectral Joint Of Hyperspectral Image Classification Based On Multi-branch 3D Convolution

Hyperspectral image(HSI)classification has become a hot topic in the field of remote sensing.Due to the complexity of hyperspectral data,it is difficult for traditional methods to accurately classify hyperspectral data.Deep learning has been recognized as a powerful feature extraction tool to effectively address nonlinear problems and is widely used in HSI classification and demonstrated good performance.However,existing deep learning methods still suffer from problems that spatial and spectral information is lost and spectral contextual information is not sufficiently extracted.In order to fully explore the intrinsic relationship between spatial and spectral information and extract more distinguishable discriminative information.In the three-dimensional convolutional neural network(3D CNN),this paper further improves the classification accuracy of hyperspectral images by combining deep learning methods such as bi-directional long and short-term memory(Bi-LSTM)network and Dense Net.The main research work is as follows:(1)To improve the utilization of the spatial and spectral information from the HSI,a framework based on three-dimensional convolutional neural network(3D CNN)and band grouping–based bidirectional long short-term memory(Bi-LSTM)is proposed in this paper.In the framework,the spectral data is regarded as sequence data,and the Bi-LSTM network acts as the spectral feature extractor of this framework to fully exploit the close relationships between spectral bands.The 3D CNN has a unique advantage in processing the 3D data,so it is used as the spatial-spectral feature extractor in this framework.Finally,in order to optimize the parameters of both feature extractors simultaneously,the Bi-LSTM and 3D CNN share a loss function to train the network.The framework obtains more accurate results on three datasets in comparison with state-of-the-art methods.(2)To avoid losing information in the feature extraction stage,a multi-branch 3D-densely connected network for HSI classification in this paper.The network can fully exploit the spatialspectral information contained in HSI and extract richer spatial-spectral features using convolutional kernels of different sizes and densely connected networks.The spatial features are subsequently enhanced using a spatial attention mechanism to provide more easily identifiable spatial features for HSI classification.A global average pooling is used instead of a fully connected layer to reduce the number of parameters of the whole network.To evaluate the performance of the proposed method,HSI classification is tested on three datasets,and the experimental results demonstrate the effectiveness of the proposed method.