Convolutional Neural Networks (CNNs) excel at capturing local features but face limitations in capturing global contextual information, especially in medical images with significant spatial variations. Although Transformers introduce relative locational encoding to improve long-range dependency modeling, their high computational complexity poses challenges, particularly in high-resolution medical images, where global information capture remains inefficient. To address these issues, this paper introduces a Location-Aware Module and a Feature Correction Module. The Location-Aware Module enhances the representation of global locational information by constructing prototypes through block partitioning and performing random location prediction. The Feature Correction Module leverages contrastive learning to correct misclassified pixels, thereby improving segmentation accuracy. Experimental results demonstrate that the proposed method achieves superior performance in multiple medical image segmentation tasks, particularly in capturing global image information in complex medical images.