Aprendizaje profundo para la detección de malezas en sistemas agrícolas y forestales mesoamericanos

Introduction. Weed control is a permanent activity in tropical agricultural and forestry systems. Deep learning offers possibilities for weed detection, identification, and management. Objective. To describe the main advances and applications of deep learning for weed identification and localization and analyze its applicability as a supporting tool for weed management in agricultural and forestry systems in Mesoamerica. Development. A scientific literature review was conducted by searching for articles published between 2016 and 2025 in Scopus, Web of Science, ScienceDirect, and SpringerLink. The review was primarily based on 15 studies that applied deep learning techniques and reported performance metrics, and 15 review articles. The components of the process were identified, and a methodological route for implementation in Mesoamerica was proposed, considering the particularities of tropical agricultural and forestry systems. Deep learning models have shown favorable performance for weed recognition in various agricultural systems. No studies in forest plantations were found in the search conducted, although potential for application during the initial stages of establishment was identified. The roadmap considers the prioritization of crops, forest species, and weeds with the greatest impact, the coordinated development of databases, and the training and implementation of models. Conclusions. Deep learning presents relevant advances for weed identification and localization. Its implementation in Mesoamerica could contribute to management through precision agriculture and forestry strategies but requires representative data, field validation, and integration with decision support systems.