At present, most semantic segmentation networks are based on Transformer framework. Transformer are capable of processing long sequences of data and have the advantage of parallel computation. However, due to the high computational complexity, it cannot be used in resource-constrained environments. In this paper, we built a lightweight network, which replaces the traditional backbone with grouped full convolution and Transformer. The grouped full convolution is a plug and play module that can be applied to different networks. It extracts features by grouping dynamic and static convolution. Different convolution has specific responses to different categories. In addition, it adds a lightweight segmentation head to improve the segmentation capability while reducing the number of parameters. Experimental results demonstrate that the proposed approach exhibits strong performance compared to several leading existing techniques.