摘要：【中文】本发明涉及一种联合3D/2D卷积网络和自适应光谱解混的高光谱图像分类方法，通过使用3D/2D稠密连接网络和多个中间分类器构建网络模型，又将自适应光谱解混作为网络分类结果的补充。具有早退出机制的多个中间分类器的设计使得模型可以使用自适应光谱解混来促进分类，这为计算量和最终分类性能带来了相当大的益处。此外，本发明还提出了一个基于空谱特征的3D/2D卷积，使得三维卷积能够包含较少的三维卷积，同时通过利用二维卷积获得更多的光谱信息来增强特征学习，从而降低了训练的复杂度。本发明与现有的基于深度学习的高光谱图像分类方法相比，计算效率更高，精度更高。 【EN】The invention relates to a hyperspectral image classification method combining a 3D/2D convolutional network and adaptive spectrum unmixing. The design of multiple intermediate classifiers with early exit mechanisms allows the model to use adaptive spectral unmixing to facilitate classification, which brings considerable benefits to computational effort and final classification performance. In addition, the invention also provides a 3D/2D convolution based on the space spectrum characteristics, so that the three-dimensional convolution can contain less three-dimensional convolution, and meanwhile, more spectrum information is obtained by utilizing the two-dimensional convolution to enhance characteristic learning, thereby reducing the training complexity. Compared with the existing hyperspectral image classification method based on deep learning, the hyperspectral image classification method based on deep learning is higher in calculation efficiency and higher in precision.