摘要
针对现有目标检测模型在玉米叶片病害检测任务中存在识别精度不足及参数量过大的问题,提出了一种基于YOLOv10n的玉米叶片病害检测改进模型。通过对比YOLOv5n,YOLOv6n,YOLOv8n,YOLOv9t等经典模型,发现YOLOv10n检测精度最高;利用离线增强技术对原始数据集进行扩充,有效提高了模型的检测精度;在骨干网络中引入CBAM注意力机制,增强了模型对重要特征的关注程度。增强后的YOLOv10n算法与原始YOLOv10n相比,准确率提升了7.3%,Recall提升了5.5%,mAP@0.5提升了10.9%。加入CBAM注意力机制的YOLOv10n-Enhance算法与原始YOLOv10n相比,准确率提升了9.5%,Recall提升了6.7%,其mAP@0.5达到了93.9%。改进后的YOLOv10n算法在不增加参数量和计算复杂度的情况下,显著提高了对玉米叶片病害的检测能力。
Aiming at the problems of insufficient recognition accuracy and excessive parameter amount of the existing target detection models in the maize leaf disease detection task,an improved model for maize leaf disease detection based on YOLOv10n is proposed.By comparing the classical models such as YOLOv5n,YOLOv6n,YO-LOv8n,and YOLOv9t,it is found that YOLOv10n has the highest detection accuracy.The original dataset is expand-ed by using offline enhancement technology,which significantly improves the detection accuracy of the model.The introduction of the CBAM attention mechanism in the backbone network enhances the model's to focus on important features.Compared with the original YOLOv10n algorithm,the enhanced YOLOv10n algorithm's accuracy has in-creased by 7.3%,Recall has increased by 5.5%,and mAP@0.5 has increased by 10.9%.Compared with the original YOLOv10n,the accuracy of YOLOv10n-Enhance algorithm by adding CBAM attention mechanism has increased by 9.5%,and Recall has increased by 6.7%,with mAP@0.5 reaching 93.9%.The improved YOLOv10n algorithm signif-icantly improves the detection of maize leaf diseases without increasing the number of parameters and computational complexity.
作者
党珊珊
白明宇
路扬
潘春宇
赵晨雨
乔世成
DANG Shanshan;BAI Mingyu;LU Yang;PAN Chunyu;ZHAO Chenyu;QIAO Shicheng(College of Computer Science and Technology,Inner Mongolia Minzu University,Tongliao 028043,China)
出处
《内蒙古民族大学学报(自然科学版)》
2025年第2期69-76,共8页
Journal of Inner Mongolia Minzu University:Natural Sciences Edition
基金
国家自然科学基金项目(62162049)
内蒙古自治区直属高校基本科研业务费项目(GXKY23Z015)
内蒙古民族大学博士科研启动基金项目(BS658)。
