文章摘要
孙丙宇,单超,房永峰.基于改进YOLOv7的番茄果实目标检测[J].安徽建筑大学学报,2024,32():
基于改进YOLOv7的番茄果实目标检测
Tomato fruit target detection based on improved YOLOv7
投稿时间:2023-05-19  修订日期:2023-07-06
DOI:
中文关键词: 目标检测  YOLOv7  注意力机制  Soft-NMS  BiFPN  
英文关键词: Object detection  YOLOv7  Attention mechanism  Soft-NMS  BiFPN
基金项目:中国科学院重点资助项目
作者单位邮编
孙丙宇 安徽建筑大学 230601
单超* 安徽建筑大学机械与电气工程学院 230601
房永峰 中国科学技术大学研究生院科学岛分院 合肥 
摘要点击次数: 140
全文下载次数: 59
中文摘要:
      针对农业采摘机器人在采摘过程中容易受到果实重叠、果实遮挡和果实体积小等一系列问题,提出一种改进YOLOv7网络对番茄果实进行目标检测。首先在YOLOv7网络结构中增加SimAM注意力模块和CA注意力模块,提高网络特征提取能力;其次特征融合网络的张量拼接操作与加权双向特征金字塔BiFPN结合,提高特征融合能力;再用Soft-NMS算法代替NMS算法,增加网络对重叠区域的检测能力;最后CIOU Loss替换成EIOU Loss,优化网络性能。实验结果表明,改进后的网络mAP值可达96.7%,准确率为96.2%,召回率为99.0%,检测时间为14.102ms,改进后的YOLOv7网络可以同时满足精度和检测速度的要求,该方法可以为番茄采摘提供技术支持。
英文摘要:
      Aiming at a series of problems in the picking process of agricultural picking robots, such as fruit overlap, fruit occlusion and small fruit size, an improved YOLOv7 network was proposed for tomato fruit target detection. Firstly, the SimAM attention module and CA attention module are added to the YOLOv7 network structure to improve the feature extraction capability. Secondly, the tensor splicing operation of feature fusion network is combined with weighted bidirectional feature pyramid BiFPN to improve the feature fusion ability. Then replace the NMS algorithm with the Soft-NMS algorithm to increase the detection capability of the network for overlapping areas. Finally, CIOU Loss is replaced with EIOU Loss to optimize network performance. Experimental results show that the improved network mAP value can reach 96.7%, the accuracy rate is 96.2%, the recall rate is 99.0%, and the detection time is 14.102ms. The improved YOLOv7 network can meet the requirements of accuracy and detection speed at the same time. The method can provide technical support for tomato picking.
查看全文   查看/发表评论  下载PDF阅读器
关闭

分享按钮