基于改进SSD的风机叶片表面缺陷检测方法

季利鹏; 吴世龙; 聂涛; 杨文威; 杨迦迤

本文已被：浏览 1160次下载 747次

中文摘要: 针对无人机对风机叶片表面缺陷检测中出现的多尺度目标问题，本文提出一种基于改进SSD的风机叶片缺陷检测方法。以具有多尺度结构框架的目标检测模型SSD为基础，引入残差网络ResNet50作为其特征提取网络，用以获取更深层次的细节特征信息，从而提升缺陷检测模型的整体效果。在建立的风机叶片表面缺陷图像数据集下进行模型验证，结果表明，该方法的平均精确度mAP@.5为84.29%，与YOLOv3和RetainNet相比，对各类型缺陷的平均精确度分别提高了2.92%和8.69%，同时较传统SSD算法平均精确度提升了2.21%。

中文关键词: 无人机风机叶片缺陷检测改进SSD 残差网络

Abstract:Aiming at the multi-scale target problem of detecting wind turbine blade surface defects by unmanned aerial vehicle (UAV), this paper proposes an improved SSD-based defect detection method for wind turbine blades. Based on the target detection model SSD with the multi-scale structural framework, the residual network ResNet50 is introduced as its feature extraction network to obtain deeper detailed feature information, thus improving the overall effect of the defect detection model. Model validation is performed under the established wind turbine blade surface defect image data set, and the results show that the mean average precision (mAP@.5) is 84.29%. Compared with YOLOv3 and RetainNet, it improves the average accuracy of each defect type by 2.92% and 8.69%, respectively. Moreover, it also improves the average accuracy by 2.21% compared with the traditional SSD algorithm.

keywords: UAV wind turbine blade fault detection improved SSD ResNet50

文章编号：20230308 中图分类号：TP391.9 文献标志码：A

基金项目:上海市 “科技创新行动计划”人工智能科技支撑专项项目(20511101600)；上海市新能源智能运维专业技术服务平台（22DZ2291800）

作者	单位
季利鹏	上海理工大学机械工程学院
吴世龙	上海理工大学机械工程学院
聂涛	上海航天控制技术研究所
杨文威	上海嘉柒智能科技有限公司
杨迦迤	上海嘉柒智能科技有限公司

Author Name	Affiliation
JI Lipeng	School of Mechanical Engineering, University of Shanghai for Science and Technology
WU Shilong	School of Mechanical Engineering, University of Shanghai for Science and Technology
NIE Tao	Shanghai Aerospace Control Technology Institute
YANG Wenwei	Shanghai JiaQi Intelligent Technology Limited Company
YANG Jiayi	Shanghai JiaQi Intelligent Technology Limited Company

引用文本：
季利鹏,吴世龙,聂涛,杨文威,杨迦迤.基于改进SSD的风机叶片表面缺陷检测方法[J].飞控与探测,2023,(3):63-71.