To address the challenges of inadequate real-time detection and limited accuracy in identifying seedlings and weeds within maize field environments, we propose a novel weed detection method based on an enhanced version of YOLOv5s. Our study focuses on maize seedlings and four commonly found associated weeds. Utilizing YOLOv5s, which integrates detection accuracy and speed, as the foundational model, we incorporate the DCA attention module proposed in this study into the C3 structure of the feature extraction network to augment the model's capacity for expressing features.In the loss function calculation, the EIOU loss function is utilized to quantify positioning loss during model training and optimize both convergence speed and positioning accuracy of the model. Experimental results demonstrate that the enhanced YOLOv5s model achieves a 95.7% mAP@0.5 and an 81.9% mAP@0.5:0.95 on the maize and associated weeds dataset, with a detection frame rate of 61 frames per second, meeting both accuracy and real-time requirements.