| 摘 要: 为了实现对水稻病害的精准检测,文章基于YOLOv8n模型(You Only Look Once version 8 nano)提出了一个全新的改进模型YOLO-Rice。该模型通过3项关键的技术创新,提升了对水稻叶片和稻穗病害的检测精度。首先模型在骨干网络中引入CBAM(Convolutional Block Attention Module)卷积注意力机制;其次模型采用Gold-YOLO的GD(Gather-and-Distribute)机制,在模型的颈部进行特征融合;最后更换了传统的损失函数,采用WIoU作为新的损失函数。通过上述改进,YOLO-Rice模型在平均精度均值(mAP50%)上实现了3.4百分点的显著提升,最终达到了96.0%的准确率,充分证明了YOLO-Rice模型在水稻病害检测任务中的有效性。 |
| 关键词: 目标检测;CBAM;Gold-YOLO;WIoU |
|
中图分类号: TP391
文献标识码: A
|
| 基金项目: 科技部国家外专项目(G2022042005L);甘肃省重点研发计划(23YFWA0013);甘肃省高等学校产业支撑项目(2023CYZC-54);兰州市人才创新创业项目(2021-RC-47);2023年甘肃农业大学美育和劳动教育教学改革项目(2023-09) |
|
| Rice Disease Detection Based on CBAM Attention and Gather-and-Distribute Mechanism |
|
ZHOU Tianxiang, WEI Linjing
|
(College of Inf ormation Science and Technology, Gansu Agricultural University, Lanzhou 730070, China)
1448368118@qq.com; 916277964@qq.com
|
| Abstract: In order to achieve precise detection of rice diseases, this paper proposes a novel improved model called YOLO-Rice, based on the YOLOv8n model (You Only Look Once version8 nano). This model enhances the detection accuracy of rice leaf and panicle diseases through three key technological innovations. Firstly, the Convolutional Block Attention Module (CBAM) is introduced into the backbone network; secondly, the Gold-YOLO's GD (Gather-and-Distribute) mechanism is employed for feature fusion at the neck of the model; finally, the traditional loss function is replaced with WIoU ( Wise Intersection over Union ), the new loss function. Through these improvements, the YOLO-Rice model achieves a significant increase of 3. 4 percentage points in mean Average Precision (mAP50% ), ultimately reaching an accuracy of 96.0% . This clearly demonstrates the effectiveness of the YOLO-Rice model in the task of rice disease detection. |
| Keywords: object detection; CBAM; Gold-YOLO; WIoU |
