Optimisation of U-Net Semantic Segmentation Model Based on Residual Connectivity and Attention Mechanisms

Abstract

In this paper, an improved U-Net semantic segmentation model based on residual connection and attention mechanism is proposed, aiming to improve the accuracy and stability of image segmentation. Adding residual connections between each convolutional layer of U-Net enables the input to bypass one or more convolutional layers and sum up with the outputs of these layers. This not only improves the training efficiency of the network, but also enhances the robustness and stability of the model. In addition, the model introduces two attention mechanisms based on U-Net: Channel Attention and Spatial Attention. Experimental results on the knee MRI dataset show that the improved model outperforms the traditional U-Net and other comparative methods in evaluation metrics such as loss rate, mean absolute error (MAE), and F1 score, demonstrating its advantages in medical image segmentation tasks. The important information in the image is captured more effectively and the segmentation accuracy is improved. It has important reference value for the medical image processing field.