Brain tumor segmentation is an important step in the diagnosis, treatment planning, and monitoring of patients with brain tumors. Efficient and accurate segmentation of tumors can greatly improve the accuracy and reliability of medical diagnosis. In this paper, a deep convolutional neural network model is designed to achieve precise automatic segmentation of brain tumor MR images. Residual-to-residual networks are used to encourage feature reuse, and an atrous convolutional pyramid pooling module is introduced to obtain multiscale information. Spatial and channel attention mechanisms are used to efficiently extract features, thereby improving the robustness of the network and achieving high-precision automatic segmentation of brain tumor MRI images. The combined loss was trained and tested on BRATS2021 data sets, and the results of different evaluation indexes in WT, TC and ET were compared with U-Net, Res-Unet, Dense-Unet and DeepLabv3 +, which proved that the network model has strong segmentation performance and can realize accurate segmentation of brain tumor MR images.