@Article{cmc.2026.086207,
AUTHOR = {Peng Shen, Tenglong Li, Yongpeng Sun, Hao Cui, Guoqing Zhang},
TITLE = {A Review of Deep Learning-Based Precision Weed Detection in Farmland Environments},
JOURNAL = {Computers, Materials \& Continua},
VOLUME = {},
YEAR = {},
NUMBER = {},
PAGES = {{pages}},
URL = {http://www.techscience.com/cmc/online/detail/27979},
ISSN = {1546-2226},
ABSTRACT = {Agriculture plays an important role in food security and social development. With the rapid development of the Fourth Agricultural Revolution, also known as Agriculture 4.0, traditional weed control methods that rely on manual experience and uniform herbicide application can no longer meet the demands for efficient, precise, and environmentally friendly production. Farmland weeds compete with crops for light, water, and nutrients, thereby seriously affecting crop yield and quality. Therefore, the development of efficient weed detection and recognition technologies is of great significance. In recent years, the rapid progress of deep learning in computer vision has provided new technical approaches for automatic weed recognition, precise localization, and intelligent weed control in farmland environments. This paper presents a review of deep learning-based precision weed detection technologies in farmland. Unlike previous reviews that mainly focus on weed recognition algorithms, datasets, or agricultural robotic systems separately, this review provides a complementary Agriculture 4.0-oriented and deployment-aware perspective by linking data foundations, deep learning models, field robustness, edge deployment, and precision weeding applications within a unified technical framework. First, mainstream weed image datasets, data preprocessing methods, and data augmentation strategies are summarized. Then, recent advances in the application of convolutional neural networks to weed image classification, object detection, semantic segmentation, and instance segmentation are analyzed. In addition, the application value of semi-supervised, unsupervised, and weakly supervised learning in small-sample scenarios is discussed, and research progress on the deployment of weed detection models on edge devices is reviewed. Finally, this paper summarizes the key challenges associated with complex field environments, visual similarity between crops and weeds, occlusion and overlap, and real-time deployment, providing a reference for the subsequent design of weed detection models and the application of precision weeding equipment.},
DOI = {10.32604/cmc.2026.086207}
}