To address the challenge of mechanical weeding, a YOLOv5‑based method was developed for detecting corn plants. A complex field environment dataset was established to train the model. The original YOLOv5 architecture was modified by introducing a Coordinates Attention (CA) mechanism in both the backbone and head layers. This mechanism dynamically enhanced the weighting of corn plant location information, thereby improving detection accuracy. Additionally, a weighted Bidirectional Feature Pyramid Network (BiFPN) was incorporated in the neck layer to strengthen feature fusion, which further enhanced both detection speed and accuracy. The test results showed that compared with the original model, the mAP@0.5 and mAP@0.5:0.95 of this method had increased by 4.31 and 3.66 percentage points respectively. The detection speed and model size were 46.77 frames/s and 15.56 M respectively. It also has certain advantages when compared with SSD, YOLOv5, Fast R—CNN and YOLOv7. From this, it can be concluded that the designed model can effectively achieve the detection of corn plants, with good real‑time performance and small memory usage, which can provide a reference for the seedling protection work of intelligent weeding robots.