改进YOLOv5s的清洁船水面漂浮垃圾识别算法  

Improved YOLOv5s Algorithm for Floating Garbage Detection by Cleaning Ships

在线阅读下载全文

作  者:周波 袁和平 刘必劲 陈水宣 ZHOU Bo;YUAN Heping;LIU Bijin;CHEN Shuixuan(School of Mechanical and Automotive Engineering,Xiamen University of Technology,Xiamen 361024,China;Fujian Key Laboratory of Green Intelligent Cleaning Technology and Equipment,Xiamen 361024,China;Hainan Institute,Zhejiang University,Sanya 572025,China)

机构地区:[1]厦门理工学院机械与汽车工程学院,福建厦门361024 [2]福建省绿色智能清洗技术与装备重点实验室,福建厦门361024 [3]浙江大学海南研究院,海南三亚572025

出  处:《厦门理工学院学报》2024年第5期40-50,共11页Journal of Xiamen University of Technology

基  金:福建省自然科学基金项目“基于水弹性理论和智能优化的海洋多浮体阵列布置”(2022J05281);海南省科技专项“海南省海洋生态灾害预警监测与防控辅助决策支持系统及示范应用”(ZDYF2023SHFZ119)。

摘  要:为解决水上漂浮垃圾种类繁多、移动端算力有限、现有识别模型较复杂等原因导致的水上清洁船识别垃圾精度低、速度慢等问题,提出一种基于YOLOv5s(you only look once version5 small)网络模型的水上漂浮垃圾识别算法。该算法采用K-means聚类算法调整边界框比例,提高检测精度;以渐进式学习方式EfficientNetv2模型替代YOLOv5s的主干部分,融合高效通道注意力(efficient channel attention,ECA)机制,减少模型复杂度,提高检测速度,同时增强模型的特征提取能力;引入平衡因子φ和归一化高斯Wasserstein距离(normalized Gaussian Wasserstein distance,NWD)度量对YOLOv5s的损失函数优化,降低模型对水上远距离漂浮垃圾的检测敏感性。自制数据集的测试实验结果显示,改进算法的mAP比YOLOv5s算法提高2.2%,模型的参数量下降20.34%,检测速度提高30.84%,表明改进算法具有优越性。To improve the low accuracy and low speed in floating garbage detection by cleaning ships due to the wide variety of floating garbage,limited computing power of mobile devices,and complexity of existing recognition models,a floating garbage recognition algorithm based on the YOLOv5s(you only look once version 5 small)network model is proposed.This algorithm adjusts the bounding box ratios using the K-means clustering algorithm to improve detection accuracy,replaces the backbone part of YOLOv5s with the EfficientNetv2 model in a progressive learning approach,integrates the efficient channel attention(ECA)mechanism to reduce model complexity,enhance detection speed and improve the model’s feature extraction capabilities,introduces a balance factorφand normalized Gaussian Wasserstein distance(NWD)metric to optimize YOLOv5s’s loss function so as to reduce the model’s sensitivity to detection of distant floating garbage on water.Test results on a custom dataset show that the improved algorithm’s mAP is increased by 2.2%compared to the YOLOv5s algorithm,the model’s parameter count reduced by 20.34%,and the detection speed increased by 30.84%,all this indicating the superiority of the improved algorithm to others.

关 键 词:清洁船 水面漂浮垃圾 识别算法 YOLOv5s算法 K⁃means聚类算法 EfficientNetv2 注意力机制 损失函数 

分 类 号:TP391.4[自动化与计算机技术—计算机应用技术]

 

参考文献:

正在载入数据...

 

二级参考文献:

正在载入数据...

 

耦合文献:

正在载入数据...

 

引证文献:

正在载入数据...

 

二级引证文献:

正在载入数据...

 

同被引文献:

正在载入数据...

 

相关期刊文献:

正在载入数据...

相关的主题
相关的作者对象
相关的机构对象