Existing pedestrian detection methods that utilize the fusion of visible and infrared modalities face issues of the high rate of missed detections in low-illumination scenarios. To address this problem, this study proposes an innovative multispectral nighttime pedestrian detection method. Firstly, we enhanced the popular low-light enhancement algorithm Zero-DCE by integrating a specially designed denoising module, which effectively improves the clarity of pedestrian targets in visible light images, providing higher quality image inputs for subsequent detection tasks. Next, we extended the existing detector framework with an infrared image processing branch, forming a dual-channel detection network for visible and infrared modalities. By introducing a carefully designed feature fusion module, we achieved deep fusion of the features from both branches, significantly enhancing the model’s feature extraction and target detection capabilities. Finally, we tested the improved model on the LLVIP and KAIST datasets, and the results show an improvement in detection performance.