Aiming at the problems of missed detection, occlusion and low accuracy of target detection methods in foggy scenes, a foggy target detection algorithm YOLO-CL-CA based on multi-scale feature fusion is proposed. Firstly, in the data pre-processing stage, the AOD-Net model is used to defog the RTTS dataset to improve the image detail information. Secondly, a centralized feature pyramid CFPNet (Centralized Feature Pyramid) is introduced to regulate the shallow features with deep features to capture the key local regions of images and enhance the image feature utilization capability of the model. Thirdly, the CA attention mechanism (Coordinate Attention) is added before the output layer to improve the model’s ability to capture small target features. Finally, the LKC3 module is constructed by combining large convolution kernel to improve the problem of missed detection due to occlusion. The experimental results show that, the accuracy and mAP0.5 of the proposed algorithm are 90.6% and 81.7% respectively, 4.2% and 1% higher than these of YOLOv5s, which proves that the improved algorithm is effective and practical for fog target detection.