<p>The details of the turbulent flow field are crucial to explore turbulent characteristics of small-scale structures. However, obtaining high-resolution turbulence fields at a low cost is a long-standing challenge in fluid mechanics studies. Therefore, we propose a super-resolution turbulence reconstruction framework based on multi-scale hybrid attention Swin-transformer (MHASTR) to reconstruct high-resolution turbulence fields from low-resolution data. The framework mainly consists of three modules: the shallow feature extraction module, the deep feature extraction module and the high-resolution field reconstruction module. The first two modules use a multi-scale feature extraction mechanism and a hybrid attention mechanism to extract turbulence features. The last module is mainly to reconstruct turbulence fields using the extracted features. Moreover, a physics-based loss function is designed to restrict the solution space and guide the operation of this model. Finally, the performance of proposed model is evaluated by the forced isotropic turbulence and turbulent channel flow datasets. Experimental results show that MHASTR can efficiently extract the deep nonlinear features, multi-scale interaction and scale-invariance features, and reconstruct high-resolution turbulence structures from low-resolution inputs. Compared with other methods, MHASTR can restore most of the physical properties from low-resolution data with only about 4.2% of the number of parameters of other state-of-the-art methods.</p>