The underwater image segmentation is a difficult task owing to challenges like light scattering, color distortion, high noise levels, and low contrast. These issues limit the performance of conventional segmentation techniques, highlighting the need for more advanced and adaptive methods specifically designed for underwater conditions. This paper proposes a method called Advanced Underwater Segmentation with ResNet50 with enhancement-guided Squeeze and Excitation Blocks (AUSREGSE), providing an effective framework for segmenting underwater scenes through the integration of image enhancement, attention mechanisms, and multi-scale feature fusion. The process starts with an enhancement branch designed to improve image quality by addressing color distortions, increasing contrast, and suppressing noise. The resulting enhanced image is then utilized to guide the Squeeze-and-Excitation (SE) blocks, which are applied to multi-level features extracted from the ResNet-50 backbone to refine channel-wise attention. Subsequently, a Feature Pyramid Network (FPN) aggregates the refined features from multiple scales, ensuring the retention of both detailed spatial information and rich semantic representations, fed to the final segmentation module generates precise pixel-wise predictions, enhancing the model’s effectiveness in handling visually degraded and complex underwater environments.