Ear detection method and system based on transfer learning and improved YOLOv11
By constructing a self-built dataset and improving the YOLOv11 model, and combining transfer learning and cascaded group attention mechanisms, the problems of missed detection and false detection in densely distributed and occluded wheat ear scenarios were solved, achieving high-precision and high-robust wheat ear detection results.
CN122200628APending Publication Date: 2026-06-12ANHUI AGRICULTURAL UNIVERSITY
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- ANHUI AGRICULTURAL UNIVERSITY
- Filing Date
- 2026-01-30
- Publication Date
- 2026-06-12
Smart Images
Figure CN122200628A_ABST
Abstract
The application discloses a wheat ear detection method and system based on transfer learning and improved YOLOv11, and relates to the technical field of computer vision agricultural target detection; the method comprises the following steps: collecting a wheat ear original image, processing the wheat ear original image to establish a wheat ear original image dataset; selecting a YOLOv11 model as a benchmark model, pre-training the YOLOv11 model based on a wheat ear public dataset, fine-tuning the YOLOv11 model by using the pre-trained weight in combination with the wheat ear original image dataset; improving the YOLOv11 model based on fine-tuning, introducing a cascade group attention mechanism into a backbone network, introducing an ASF-YOLO module into a neck network, and obtaining a target detection network model after improvement; and training and verifying the model. The application is suitable for detecting wheat ears in a natural field environment, and aims to solve the problem of detecting and identifying wheat ears in a dense, occluded and complex background while ensuring high recognition accuracy with a self-built small dataset.
Need to check novelty before this filing date? Find Prior Art