As pressure on global food security mounts, controlled-environment agriculture (CEA) demands innovations in monitoring and control. This review surveys computer-vision methods for greenhouse pests and diseases, covering foundations, core techniques, and applications, highlighting early detection, multimodal fusion, and decision support. We trace the evolution from traditional machine vision to AI-driven methods, spanning image acquisition and preprocessing, feature extraction, and semantic understanding. Deep-learning-based pest and disease recognition improves accuracy under controlled conditions, and enables earlier detection of small-target pests and incipient diseases. Multimodal fusion and transfer learning further improve model generalization in complex greenhouse scenes. We also explore computer vision-driven intelligent decision-making systems, including disease and pest assessment, precision intervention, and optimization of integrated prevention and control strategies. Case studies indicate that AI-based vision systems can provide pest and disease warnings several days in advance, which reduces pesticide use and improves production efficiency. However, bottlenecks such as insufficient environmental adaptability, difficulties in learning from small samples, and limited computational resources still hinder broader implementation. Future research should prioritize multisource heterogeneous data fusion, lightweight model design, and the integration of agronomic knowledge. Together, these directions can support monitoring from phenotypes to mechanisms and from individual plants to populations.
Yu et al. (Fri,) studied this question.