MS-2HCNN:Hyperspectral Imagery Signal Classification Method Based on Deep Learning
In order to extract and combine space in hyperspectral imagery signals more accurately,an MS-2HCNN structure(Multi Stage-Heightened&Hyperspectral convolutional neural network(MS-2HCNN)is proposed,which obtains more discriminative features by com-bining the results of different convolutional layers,generates a large number of features after dynamic merging,connects the extracted spectral and spatial information in series,simplifies the calculation,and ensures the accuracy and reliable classification performance.The method improves the classification performance by using existing feature combination techniques.In addition,for the proposed multi-stage design,the background information obtained by the upper layer can be combined with the precise spatial information ob-tained by the lower layer,makes it have more advantages than the existing methods in terms of accuracy and complexity.Finally,network optimizer with better complexity and precision is introduced,and batch normalization is adopted,reducing the number of parameters and improving the fitting ability of the MS-ZHENN model.The clussification results on different open source datasets show the validity of the proposed method.
万方数据
维普
