Multimodal tomato disease dataset

Tomato leaf diseases are a significant factor affecting crop yield and quality in protected agriculture, and their accurate identification and scientific control are crucial for the development of smart agriculture. Addressing the limitations of existing crop disease datasets, which primarily rely on single image or category annotations and lack semantic information and fine-grained perception annotations, this paper constructs a multimodal dataset for tomato leaf diseases. The dataset was collected in real greenhouse environments and covers six typical disease categories, including tomato gray leaf spot, tomato leaf miner larval damage, tomato late blight, tomato bacterial spot, tomato leaf mold, and tomato pesticide damage. It contains 546 original images, which are expanded to 1,638 images through data augmentation. According to the manifestation characteristics of different diseases, the dataset provides object detection annotations for both local lesion regions and overall disease areas. In addition, each image is accompanied by structured textual annotations that include symptom descriptions and corresponding control recommendations. The data quality is validated through cross-modal retrieval and object detection experiments, and the results demonstrate that the proposed dataset can effectively support research on tomato leaf disease detection, vision–language multimodal modeling, and agricultural intelligent question answering. Overall, the dataset provides a reliable data foundation for intelligent perception and precise management of tomato leaf diseases.