A Multimodal Image-Text Dataset for Potted Vegetables

The potted planting model, with advantages such as controllable environment and resource intensiveness, is widely used in facility vegetable production. However, its key operational stages still heavily rely on manual labor, urgently necessitating the introduction of intelligent technologies for precise management and control. Most current methods for recognizing the growth status of facility vegetables are based on unimodal visible light images, which are prone to interference from light variations, plant occlusion, and substrate complexities in the intricate potted planting environment, making it difficult to adapt to practical production scenarios. Furthermore, existing agricultural datasets are predominantly focused on field or orchard settings, lacking multimodal image-text data resources specifically for potted vegetables, which constrains the development and application of multimodal perception technologies and agricultural large models in this domain. To address this, this study constructed a multimodal image-text dataset covering eight common types of potted vegetables across three growth stages, 185.36 GB in total. The image data includes three modalities: visible light, depth, and near-infrared, supporting tasks such as object detection, instance segmentation, and multimodal visual fusion to meet the recognition requirements for intelligent agricultural machinery operations. The text data comprises two types: scene semantic descriptions and agricultural knowledge question-answering, which can facilitate the training and development of agricultural multimodal large models within potted vegetable scenarios.