Artificial Intelligence for Weed Detection: A Comprehensive Review

Weed invasion is a major constraint for agricultural production worldwide, leading to significant crop losses, increased production costs and over-reliance on herbicides. Manual inspection and blanket spraying of herbicides for weed control are labor intensive, environmentally damaging, and economically inefficient. Recent advances in artificial intelligence (AI), computer vision and precision agriculture have led to automated weed detection systems that are able to accurately identify weed species across a wide range of field conditions. Good progress is being made in weed recognition accuracy, in reducing computational complexity and in reducing herbicide use with machine learning, deep learning, object detection, semantic segmentation and vision transformer models. In addition, the application of unmanned aerial vehicles (UAVs), edge computing, Internet of Things (IoT) devices, and autonomous agricultural robots has enabled the deployment of intelligent precision farming systems. In this paper, we present a comprehensive survey on AI-based weed detection techniques, including publicly available datasets, machine learning algorithms, convolutional neural networks, object detection methods, semantic segmentation architectures, vision transformers, and emerging edge intelligence technologies. Critical discussions on the current research challenges such as dataset limitations, model generalization, computational constraints, and real-time deployment. Finally, future research directions on foundation models, explainable AI, federated learning, and autonomous precision agriculture are outlined to boost the development of sustainable weed management systems.