基于多模态文本特征融合的YOLOv11水稻病害检测OA

To address the limitations of traditional pure visual detection models,including insufficient extraction of similar rice disease features in complex backgrounds and frequent false detection and confusion,this paper explores the introduction of multimodal semantic prior knowledge to improve detection accuracy.This paper proposes a multimodal feature fusion rice disease detection model AgriText-YOLOv11,which combines the LLaVA large language model and YOLOv11.LLaVA generates exclusive natural language description texts for each rice disease image,converts them into embedding vectors,and performs cross-modal network fusion with high-resolution YOLOv11 visual features,enabling the network to focus on key discriminative regions under the guidance of text features.The proposed AgriText-YOLOv11 rice disease detection model increases the parameter quantity by only 5.2％,while improving mAP@0.5 by 1.2 percentage points and recall rate by 2.2 percentage points compared with the baseline model.Its comprehensive performance is superior to mainstream detection models such as YOLOv9,YOLOv10 and RT-DETR,which can realize efficient and accurate localization and classification of rice diseases.