YOLO-Rice:一种基于YOLOv5的水稻虫害检测  

YOLO-Rice:a Rice Pest Detection Based on YOLOv5

作　　者：巨志勇[1] 易成 周重臣 祁子翔 JU Zhiyong;YI Cheng;ZHOU Zhongchen;QI Zixiang(School of Optical-Electrical and Computer Engineering,University of Shanghai for Science and Technology,Shanghai 200093,China)

机构地区：[1]上海理工大学光电信息与计算机工程学院,上海200093

出　　处：《控制工程》2024年第12期2196-2205,共10页Control Engineering of China

基　　金：国家自然科学基金资助项目(81101116)。

摘　　要：为了保证水稻的高产和稳产、提高水稻虫害检测的识别率,提出一种高效快速的检测算法——YOLO-Rice。首先,在传统YOLOv5s模型的基础上融合CA机制;其次,更换PANet为BiFPN结构;随后,使用更加轻量的CARAFE结构来代替原始的上采样模块;最后,替换损失函数来提高收敛速度,以便更好地进行定位。结果表明,与Faster R-CNN、SSD、YOLOv3等模型相比,YOLO-Rice的检测精确率提高了2.4%~40.8%,召回率提高了6.9%~31.4%,平均精度均值提高了7%~37.2%。由此可说明,改进后的模型在多个方面都优于原模型,具有更强的鲁棒性和更高的准确率,实现了对水稻虫害的准确、高效识别。In order to ensure high and stable yield of rice and improve the recognition rate of rice pest detection,an efficient and fast detection algorithm YOLO Rice is proposed.Firstly,by integrating CA attention mechanism on the basis of traditional YOLOv5s model;Secondly,replace PANet with BiFPN structure;Replace the original upsampling module with a lighter CARAFE structure;Finally,replace the loss function to improve convergence speed and better localization.The results show that YOLO Rice improves detection accuracy by 2.4%~40.8%,recall by 6.9%~31.4%,and average accuracy by 7%~37.2% compared to Faster R-CNN,SSD,YOLOv3,and other models.The results indicate that the improved model outperforms the original model in multiple aspects,with stronger robustness and higher accuracy,achieving accurate and efficient identification of rice pest infestations.

关 键 词：目标检测 深度学习 YOLOv5 水稻虫害识别 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]