This study proposes an improved YOLOv7 model, CSCW-YOLOv7, for the detection and recognition of five types of weeds in complex wheat fields. The model incorporates several advanced techniques to enhance performance, including the CARAFE operator for up-sampling, the SE attention network for feature enhancement, the CoT module for global information capture, and the WIoU loss function for improved bounding box prediction. The model was trained on a dataset of 2,614 images containing five common weed species: Descurainia sophia, thistle, golden saxifrage, shepherd's purse herb, and Artemisia argyi. The dataset was augmented using various techniques to improve generalization and reduce imbalance among weed species. The CSCW-YOLOv7 model achieved high accuracy, recall, and mean average precision (mAP) values of 97.7%, 98%, and 94.4%, respectively, outperforming the baseline YOLOv7 model. The model also demonstrated improved efficiency, with a 10.7% reduction in parameters and a 10% decrease in floating-point operations per second (FLOPs). The model's performance was validated through ablation studies, which confirmed the effectiveness of the introduced components in enhancing weed detection. The model was compared with other deep learning models, including Faster RCNN, YOLOv5m, and YOLOv7. The CSCW-YOLOv7 outperformed these models in terms of precision, recall, and mAP, while also achieving lower computational costs. The model's ability to detect small-scale and occluded weeds was particularly notable, with high accuracy in identifying weeds in complex field environments. The study highlights the effectiveness of the improved YOLOv7 architecture in addressing the challenges of weed detection in wheat fields, including occlusion, overlapping, and varying weed sizes. The model's performance was further validated through visualizations and confusion matrices, which demonstrated its strong ability to recognize weed species and minimize false positives and false negatives. The results indicate that the CSCW-YOLOv7 is a promising tool for weed detection in wheat fields and has significant potential for practical applications.This study proposes an improved YOLOv7 model, CSCW-YOLOv7, for the detection and recognition of five types of weeds in complex wheat fields. The model incorporates several advanced techniques to enhance performance, including the CARAFE operator for up-sampling, the SE attention network for feature enhancement, the CoT module for global information capture, and the WIoU loss function for improved bounding box prediction. The model was trained on a dataset of 2,614 images containing five common weed species: Descurainia sophia, thistle, golden saxifrage, shepherd's purse herb, and Artemisia argyi. The dataset was augmented using various techniques to improve generalization and reduce imbalance among weed species. The CSCW-YOLOv7 model achieved high accuracy, recall, and mean average precision (mAP) values of 97.7%, 98%, and 94.4%, respectively, outperforming the baseline YOLOv7 model. The model also demonstrated improved efficiency, with a 10.7% reduction in parameters and a 10% decrease in floating-point operations per second (FLOPs). The model's performance was validated through ablation studies, which confirmed the effectiveness of the introduced components in enhancing weed detection. The model was compared with other deep learning models, including Faster RCNN, YOLOv5m, and YOLOv7. The CSCW-YOLOv7 outperformed these models in terms of precision, recall, and mAP, while also achieving lower computational costs. The model's ability to detect small-scale and occluded weeds was particularly notable, with high accuracy in identifying weeds in complex field environments. The study highlights the effectiveness of the improved YOLOv7 architecture in addressing the challenges of weed detection in wheat fields, including occlusion, overlapping, and varying weed sizes. The model's performance was further validated through visualizations and confusion matrices, which demonstrated its strong ability to recognize weed species and minimize false positives and false negatives. The results indicate that the CSCW-YOLOv7 is a promising tool for weed detection in wheat fields and has significant potential for practical applications.