Image super-resolution (SR) technology aims to enhance the resolution and improve the quality of images, and it has been widely used in face recognition, small target detection, medical imaging and remote sensing image analysis. Image quality assessment (IQA) is important for optimizing SR algorithms. At present, the main challenge lies in how to comprehensively learn features to characterize perceptual properties of human visual characteristics. Therefore, in this paper, we propose a no-reference quality assessment method for SR images based on local and global features fusion. First, a two-branch feature extractor is proposed, which uses convolutional neural network and vision transformer respectively to extract local features and global features. Then, considering the perception properties of the human visual system (HVS), local and global features are fused and adaptive weight strategy is applied to predict the quality score. Finally, Experimental results show that the proposed method outperforms the state-of-the-art methods in terms of prediction accuracy and generalization capability on benchmark datasets.