An enhancement-first, detection-later technical method was proposed to tackle the problem of reduced contrast caused by indistinct image details in low-light scenarios. A detail enhancement module was put forward to capture the target details in images and improve the image contrast. A multi-scale feature extraction network was constructed to fully capture the detailed information of images in the shallow layers of the network through the low-level feature extraction module and the feature fusion strategy. A similarity matching network was designed, which performed multi-scale segmentation and feature similarity matching on the original and enhanced feature maps, weighted the importance of key information, enhanced the representation of effective features and suppressed the redundant noise. When combined with the YOLOv12 target detector, the proposed method achieved the mean average precision of 83.4% and 80.4% on the nighttime target detection dataset ExDark and the self-built dataset, respectively, which significantly outperformed the existing mainstream algorithms such as PE-YOLO. The generalizability of the proposed method was validated by the results of comparative experiments conducted on the COCO and the DarkFace datasets. By enhancing the detailed features and contrast of images, the proposed algorithm improves the performance of target detection models in low-light scenarios.