Affected by atmospheric scattering, foggy images often suffer from low contrast and detail loss, resulting in degraded feature discrimination capability of conventional object detectors. To address this issue, an improved YOLOv11n-based object detection algorithm for foggy scenes, termed RGL-YOLO, is proposed. First, Receptive Field Attention Convolution (RFAConv) is introduced into the backbone network to adaptively reconstruct spatial feature weights and enhance the representation of blurred edges and critical details. Second, a Guide-Converse Feature Pyramid Network (GCFPN) is designed. In this network, the Converse2D operator is employed to perform approximate reconstruction in the frequency domain, thereby improving the preservation of high-frequency information during feature downsampling. Meanwhile, the Content-Guided Attention Fusion (CGAFusion) mechanism is incorporated to dynamically suppress background fog noise. Finally, a large-kernel selective shared head (LSK-SharedHead) is constructed to improve the discrimination and localization performance of blurred targets under low-contrast conditions. Experimental results show that the proposed method achieves mAP50 and mAP50-95 values of 73.5% and 50.0%, respectively, on the RTTS dataset, representing improvements of 4.1 and 3.7 percentage points over the baseline YOLOv11n model. Furthermore, generalization experiments conducted on the Foggy Cityscapes dataset demonstrate that the proposed method maintains favorable robustness under different fog densities, further validating its effectiveness for foggy object detection.