Medical imaging has emerged as an indispensable tool for diagnosing and treating diseases. Medical imaging facilitates the generation of images of various body parts as needed. The utilisation of medical imaging techniques has yielded improved outcomes in diagnosing diseases. Endoscopy represents a form of medical imaging that enables a doctor to inspect a patient’s body visually. Although endoscopy is primarily employed for analyzing digestive tract issues and symptoms, it is occasionally used to treat problems. Analyzing the vast footage captured during Wireless Capsule Endoscopy (WCE) procedures remains time consuming and prone to human error. In this work, we propose a novel approach to improve bleeding detection in WCE footage by integrating Duck module with axial attention mechanisms. DuckNet, known for its robust feature extraction capabilities, is augmented with co-axial attentive mechanism to enhance context understanding, thereby enhancing model performance across various tasks. The architecture integrates U-Net’s encoder and decoder formulation with DuckNet’s features, culminating in a comprehensive segmentation model. Additionally, we employ a Channel-wise Feature Pyramid (CFP) module for hierarchical feature fusion and Reverse Axial Attention for localized saliency enhancement and enriching localization information and multiscale features. We have exhaustively experimented with our proposed technique on benchmark datasets provided by MISAHUB Auto WCEBleedGEN challenge, CVC-Clinic and the Kvasir-SEG datasets and achieved competitive results.