Weed mapping using UAV imagery and AI techniques: current trends and challenges

Abstract Despite achieving accuracy rates of over 90% in recognizing weeds in crop fields using images captured by unmanned aerial vehicles (UAVs), challenges remain with embedded systems that perform automatic weed identification in real‐time. The primary objective of this review is to analyze the latest academic research on the application of machine learning/deep learning (DL) techniques for weed recognition, highlight the methodology employed, and identify the challenges encountered. A systematic review was conducted, and the retrieved papers were organized according to the strategy adopted in the proposed method. Then, for each niche, the studies were described and compared in terms of the methodology and the type of issues addressed. This review specifically covers research associated with weed mapping from images captured using UAVs, providing an in‐depth and detailed analysis of them, presenting their advantages and limitations. Regarding classical methodologies, numerous works have focused on the proposition and analysis of features aimed at extracting information related to spectral reflectance, texture, geometry, and other spatial patterns, with the goal of improving the classifier's discrimination capacity. Here, a tendency was observed with respect to non‐visible spectral channels. In contrast, DL methods stand out for their ability to extract multi‐scale features directly from images, leading to promising results in distinguishing between weed species or types. This review outlines the current landscape of UAV imagery‐based weed mapping systems, offering valuable insights for researchers and guiding future efforts toward real‐time weed mapping and the development of intelligent systems for site‐specific herbicide applications. © 2025 Society of Chemical Industry.