AI in the Field: A Review of Deep Learning Methods for Weed Identification in Wheat Crops

Weed infestation is a major constraint in wheat production, causing yield losses and higher herbicide dependence. Traditional control methods often lack precision, highlighting the need for intelligent, sustainable solutions. Deep learning has recently emerged as a powerful tool for automated and accurate weed detection in precision agriculture. This review summarizes the latest advances in deep learning applied to wheat weed identification, emphasizing model architectures, datasets, and imaging techniques. Approaches such as YOLO variants, Faster R-CNN, U-Net, and transformer-based models have achieved high accuracy in distinguishing wheat from diverse weed species, even under complex field conditions. Integration of UAV imagery, multispectral sensors, and spectral indices further enhances detection at early growth stages. Recent innovations, including attention mechanisms, feature fusion, optimized loss functions, and lightweight designs, have improved precision, speed, and generalization. Key challenges remain in dataset quality, class imbalance, and cross-field applicability. This work outlines current trends, identifies gaps, and highlights future directions for scalable and sustainable deep learning-based weed detection in wheat agriculture.