基于改进YOLOv5s的香蕉树长势研究  

Research on banana tree growth based on improved YOLOv5s

作　　者：罗明涛 米佳豪 蒋权 LUO Mingtao;MI Jiahao;JIANG Quan(Guangxi Minzu University,Nanning 530000,China;School of Artificial Intelligence,Guangxi Minzu University,Nanning 530006,China)

机构地区：[1]广西民族大学,南宁530000 [2]广西民族大学人工智能学院,南宁530006

出　　处：《计算机应用文摘》2023年第21期88-91,共4页Chinese Journal of Computer Application

摘　　要：为了提高香蕉长势的检测精准度以及检出率,文章参照香蕉的生长特性提出了一种基于改进YOLOv5s的香蕉长势研究方法,将处于生长发育时期内的三个主要时期的香蕉作为研究对象,将YOLOv5s目标检测算法作为基础构建模型,引入simAM无参注意力机制,采用SIoU替换YOLOv5s原损失函数,轻量化处理C3模块,将原模型卷积替换为MPconv,削减20*20的大目标检测层,引入BiFPN特征融合。经文章制备的香蕉树数据集验证,相较于基础YOLOv5s模型,改进后的SMBi-YOLOv5s模型的mAP提高了0.4%,模型参数量降低了76.6%,准确率提升了2.1%。In order to improve the accuracy and detection rate of banana growth detection,this article proposes a banana growth research method based on improved YOLOv5s,referring to the growth characteristics of bananas.Bananas in the three main stages of growth and development are studied,and the YOLOv5s target detection algorithm is used as the foundation to construct a model.The simAM non parametric attention mechanism is introduced,and SIoU is used to replace the original loss function of YOLOv5s,and the C3 module is lightweight processed,Replace the original model convolution with MPconv,reduce the 20*20 large target detection layer,and introduce BiFPN feature fusion.After the validation of the banana tree dataset prepared in the article,the improved SMBi-YOLOv5s model improved mAP by 0.4 percentage points compared to the basic model YOLOv5s,reduced model parameter count by 76.6%,and improved accuracy by 2.1%.

关 键 词：香蕉长势 目标检测 注意力机制 SMBi-YOLOv5s 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]