Improved U-Net with Gray Channel Attention for Image Segmentation

Image segmentation base on deep learning methods is an important direction in computer vision field. However, these models over-rely on color features in image segmentation tasks, which leads to poor segmentation effect in scenes with the interference of similar background colors. To solve this problem, this paper successfully improves the U-Net model by introducing the technical means of combining gray channel and attention mechanism. The experimental results show that compared with the original U-Net model, the average accuracy of the improved U-Net with gray channel attention has increased from 81.69% to 82.61%. At the same time, we apply this method mechanism to improved models of U-Net such as Attention U-Net and R2U-net, and similar effect is verified. These results verify that the combination of gray channel and attention mechanism can effectively improve the robustness and accuracy of deep learning model when processing color-similar background in image segmentation tasks. This work has important practical application value and provides a new solution for image segmentation tasks in complex scenes.