Abstract: This paper introduced the loss function of cluster analysis into the deep transform learning framework, proposed a deep transform cluster analysis model, and used the alternating multiplier algorithm to solve the constructed model in parallel. The deep transform clustering model selected bands based on spatial segmentation in hyperspectral data, and used the table lookup strategy to ensure the robustness of cluster center. Experiments were conducted based on Indian Pines dataset and Pavia university dataset. Compared with other end-to-end deep learning methods, the proposed method has the best operating efficiency under different band numbers, and has better overall accuracy, average accuracy and Kappa coefficient.