【摘要】 基于深度学习架构的高光谱图像分类近年来一直是遥感领域研究的热点之一.然而，如何提出新的分类框架，对具有少量标签样本的高光谱数据进行有效分类仍是一个挑战性的问题.设计了一种改进伪孪生网络的高光谱图像分类架构.该方法首先将一幅高维的高光谱图像划分为2幅低维的图像，分别利用卷积神经网络和图卷积网络进行特征提取.然后通过级联操作，将提取到的谱信息进行有效集成.最后输入全连接神经网络进行分类.所提出的方法改进了经典的伪孪生网络并应用于高光谱图像分类.在2个实际的高光谱数据集上的实验结果和比较结果验证了方法的有效性.更多还原

【Abstract】 Hyperspectral image classification based on deep learning architecture has been one of the research hotspots in remote sensing in recent years. However, it is still a challenging problem how to develop new classification frameworks to effectively classify hyperspectral data with a small number of labeled samples. To address this issue, this paper designs an improved pseudo-siamese network for hyperspectral image spectral-spatial classification. The method first divides a high-dimensional hyperspectral image into two low-dimensional images, and adopts convolutional neural network and graph convolutional network for feature extraction, respectively. The extracted spectral information is then integrated through cascade operation. Finally, the post-cascade features are input to the fully connected neural network for classification. The proposed method improves the classical pseudo-siamese network and applies it to hyperspectral image classification. Experimental results and comparative results on two practical hyperspectral datasets verify the effectiveness of the proposed method.更多还原