Global and pyramid convolutional neural network with hybrid attention mechanism for hyperspectral image classification

Convolutional neural networks (CNNs) have shown impressive results in the hyperspectral image (HSI) classification. However, they still face certain limitations that can impact their effectiveness. The kernels standard convolution fixed spatial sizes and spectral depths, which cannot capture global semantic features from HSI existing methods for extracting multiscale information are inadequate. Based on these, we propose a pyramid convolutional network with hybrid attention mechanism (GPHANet) GPHANet adopts two-branch architecture to extract both local features, leveraging strengths of each enhance classification performance. branch employs dynamic convolution, customizes parameters multiple scales based input image. This design enables model accurately comprehensively feature information, capturing subtle variations HSI. utilizes circular receptive field, enabling position layers gather entire space. allows learn contextual enhancing its understanding overall structure semantics After deep apply compact (HAM) long-range dependencies along dimensions, resulting more discriminative representation. On Pavia University, Salinas, Hong Hu datasets, proposed achieved accuracies 99.22%, 98.74%, 96.80%, respectively, using limited training samples. These much better compared state-of-the-art methods.