【摘要】 在基于深度学习的高光谱图像分类算法研究中，一些网络使用二维卷积，一些网络使用三维卷积，这两种算法都各有其优势和不足。针对基于深度学习的高光谱图像分类算法不足，提出一种融合注意力机制与混合CNN模型的算法，将高光谱图像中的空间和光谱的互补信息分别以三维卷积层和二维卷积层组合到一起。本文方法相比于经典SVM算法和目前几种性能优异的算法，在分类性能OA、AA以及Kappa系数上均有提升，证明了本算法通过结合二维卷积和三维卷积，并引入注意力机制的方法，对高光谱图像分类精度提升的有效性。更多还原

【Abstract】 In the research of hyperspectral image classification algorithms based on deep learning, some networks use 2 D-CNN, and some networks use 3 D-CNN. Both algorithms have their own advantages and shortcomings. Aiming at the shortcomings, this paper proposes an algorithm that combines the spatial information and the spectrum information in the hyperspectral image with 3 D-CNN and 2 D-CNN, and introduces an attention mechanism. Compared with the classic SVM algorithm and several excellent algorithms at present, the classification performance of the proposed method has improved in terms of OA, AA and Kappa coefficients, which proves that this algorithm combining 2 D-CNN, 3 D-CNN and attention mechanism is effective in improving the accuracy of hyperspectral image classification.更多还原