Weed management is a critical challenge in agriculture, as uncontrolled weed growth can significantly reduce crop yields and increase production costs. This study demonstrates the application of YOLO (You Only Look Once), a highly efficient and accurate real-time object detection framework. The proposed method uses the CottonWeedDet3 dataset, comprising 848 images of three weed classes—carpetweed, morningglory, and palmer amaranth—for benchmarking. In our weed detection pipeline, we observed that YOLO11 may suppress overlapping bounding boxes with low confidence scores, potentially leading to false negatives. To address this, we integrated a post-processing layer known as Soft-NMS (Soft Non-Maximum Suppression) into the pipeline, applied after the YOLO11 model’s training phase. Unlike traditional NMS, Soft-NMS reduces the confidence scores of overlapping boxes instead of discarding them outright. The YOLO11 model achieved an overall mean Average Precision (mAP) of 0.969, with class-wise mAP scores of 0.949 for carpetweed, 0.970 for morningglory, and 0.989 for palmer amaranth. In comparison, YOLOv10 achieved an mAP of 0.916 on the same dataset, highlighting the performance improvements of the proposed method. Future work will focus on replacing bottleneck modules with Dilation-wise Residual Modules (DWR) and adding Multi-Scale Modules (MSBlock) to improve multi-scale feature capture and contextual understanding.