Most existing methods for low-light image enhancement (LLIE) fail to consider the structural details in the images, which leads to problems such as blurring and distortion in the enhanced images. To address this issue, we propose a novel framework for detail enhancement termed WEDNet. Specifically, WEDNet employs two core components for detail and texture enhancement: 1) A multi-scale feature fusion (MSFF) component is used to maintain the structural integrity during the enhancement process. This component employs multi-branch parallel convolution technique, which uses different scale convolution kernels on different branches to capture and fuse the image’s multiscale structural features. 2) A bidirectional wavelet transform attention (BWTA) component, which decomposes the image into different frequencies by bidirectional wavelet transform, and then preserves important details and texture information while suppressing noise, guided by spatial attention mechanism and channel attention mechanism. Extensive experiments on eight real-world datasets show that the enhanced images of our WEDNet are more natural and detailed, and achieve superior performance compared with state-of-the-art approaches. The code is available at https://github.com/Xingshi-tech/WEDNet .