Research on the SEGDC-UNet electron microscope image segmentation algorithm based on channel attention mechanism

In this paper, we propose SEGDC-UNet, a segmentation algorithm for electron microscope (EM) images based on the channel attention mechanism. This algorithm integrates the channel attention mechanism and the GELU activation function into the DC-UNet network. By leveraging global information to selectively enhance primary features and suppress less relevant ones, the channel attention mechanism improves focus on important image channels and characteristics. Additionally, the GELU activation function enhances training performance and convergence speed. To evaluate its effectiveness, we compar SEGDC-UNet with six major lightweight image segmentation models on EMPS-Augmented electron microscopy image dataset. Experimental results demonstrate that the SEGDC-UNet model achieves higher Dice coefficient, IoU, Pixel Accuracy and Recall in electron microscopy image segmentation.