In response to the limitations of conventional single-modal object detection algorithms in complex scenarios, an infrared and visible light fusion object detection algorithm based on YOLOv7 is proposed in this paper. Firstly, a dual-branch backbone network is constructed to facilitate the extraction of feature information from both modal images. Secondly, a dual-modal feature fusion module is designed to optimally fuse features from the two different modalities, achieving effective information complementarities. Furthermore, a novel feature extraction fusion network is developed to extract and comprehensively integrate multi-scale feature layer information, enhancing inter-layer information flow and improving multi-scale detection capability. Experimental results demonstrate that the improved algorithm in this paper outperforms the original algorithm in both infrared and visible light modal scenarios, significantly enhancing detection performance in complex scenarios.