To address the problems of inaccurate lesion localization, missing boundary details, and small polyp omission in colorectal polyp segmentation, a dual-encoder network with feature-gated fusion and wavelet-enhanced modules was proposed. A dual-encoder architecture based on PVTv2 and Hiera was adopted to enhance global semantic awareness and edge representation. A feature-gated fusion module was designed to filter and combine the multi-level features extracted by two encoders adaptively, thereby better distinguishing polyp regions. A wavelet-enhanced injection module was introduced to decompose high-frequency and low-frequency information using wavelet transform, thus strengthening the edge texture expression and reduced noise, supporting fine-grained feature learning. A multi-scale prediction module was introduced by combining global average pooling and adaptive weighted fusion to achieve refined segmentation and improve adaptability to polyps with diverse shapes. Results show that the proposed network provides effective support for computer-aided diagnosis of colorectal polyps.