【摘要】 在高光谱图像分类任务中，卷积神经网络和Transformer模型优秀的特征提取能力受到了广泛关注。但主流分类模型采用简单联合空间–光谱特征进行分类，忽视了两类特征之间内在的联系和差异。为此提出了一种空间–光谱双分支注意力交互编码（double branch attention interaction encoder,DBAIE）模型。DBAIE利用空间–光谱双分支注意力机制关注重要的空间–光谱特征，并进行空间–光谱特征之间信息交互融合，融合后的特征在空间与光谱维度上实现了信息的互补与强化，从而显著增强了模型对不同地物类别的判别能力。在Indian Pines、Pavia University和Botswana数据集上，其总体分类精度分别为99.44%、99.55%和98.25%。更多还原

【Abstract】 In hyperspectral image(HSI) classification tasks, convolutional neural network(CNN)and Transformer have attracted wide attention due to their superior feature extraction capabilities.However, the current mainstream classification models use simple joint spatial-spectral features for classification, ignoring the intrinsic connections and differences between the two types of features.In this paper, we propose a spatial-spectral double branch attention interaction encoder(DBAIE)model for hyperspectral image classification. DBAIE uses a spatial-spectral dual-branch attention mechanism to focus on crucial spatial-spectral features, and performs information interaction and fusion between these two types of features. The fused features achieve information complementarity and enhancement in both spatial and spectral dimensions, thereby significantly improving the discriminative ability of the model for different ground object categories. The overall classification accuracy reaches 99.44%, 99.55%, and 98.25%, respectively.更多还原