 This paper proposes an efficient lightweight attention network architecture search algorithm, ELNAS, to automatically design a lightweight deep learning structure for hyperspectral image classification on mobile and embedded devices. ELNAS utilizes a differentiable network architecture search, NAS, with a lightweight attention module search space, and employs an edge decision strategy to avoid performance collapse caused by numerous skip operations. Experiments on real-world hyperspectral images demonstrate the effectiveness of ELNAS in achieving high-accuracy performance, with smaller parameter sizes and efficient search procedures. This article was authored by Jianing Wang, Zhenyu Hu, Yichen Lu, and others.