Punctuation restoration is a crucial post-processing step in Automatic Speech Recognition (ASR) systems, essential for improving the readability of transcribed text and facilitating the execution of Natural Language Processing (NLP) tasks. Currently, research on punctuation restoration is predominantly focused on English, with a significant scarcity of studies in low-resource languages, particularly Uyghur. To fill this academic void, we construct the Uyghur punctuation restoration dataset, UyPunc, and propose a Position Awareness (PA) module to acquire potential punctuation position features, and then integrate them with contextual features. Subsequently, we utilize a Label Attention Network (LAN) to capture punctuation information in the fused features, further enhancing the model’s punctuation prediction capability. Experimental results demonstrates the effectiveness of our approach, with an overall F1 score improvement of 8.2% on the ASR test set in our dataset compared to the baseline model. We have released our dataset on https://github.com/huangya12/UyPunc to facilitate future research on this topic.