基于改进YOLOv5x算法的海上风机叶片缺陷检测系统

随着海上风电的蓬勃发展，运维工作越来越成为突出问题。风电叶片作为风电大尺寸关键构件，其运维对机组至关重要。本文针对海上风机叶片人工运维检测存在的高风险、低效率和低精度等问题，提出了一种基于改进YOLOv5x（You Only Look Once version 5x）算法的海上风机叶片缺陷机器视觉检测系统。该方法引入了卷积块注意力机制（Convolutional Block Attention Module， CBAM），以增强神经网络对输入特征的感知能力，使用智慧交并比（Wise Intersection over Union， WIoU）作为损失函数，减少人工标注数据的误差，提高目标检测的准确性。基于海上风机叶片缺陷数据对模型进行训练，将训练好的模型封装成海上风机叶片机器视觉识别系统。试验结果显示，改进后的YOLOv5x 算法，相比于原有的YOLOv5x，平均精度均值（mean Average Precision， mAP）提高了4.71%，准确率（Precision）提高了7.48%，且能满足实时性需求。

英文摘要:

With the booming development of offshore wind power, operation and maintenance (O&M) work has become an increasingly prominent issue. Turbine blades’operation and maintenance work is crucial to the unit as a large-scale key wind power component. In order to address the problems of high risk, poor efficiency, and low accuracy in offshore wind turbine blade manual operation and maintenance inspection. This paper proposes a machine vision inspection system for offshore wind turbine blade defects, based on the enhanced YOLOv5x algorithm. In this system, a CBAM attention mechanism is introduced to enhance the neural network’s ability to perceive input features. Additionally, the Weighted Intersection over Union (WIoU) is employed as the loss function to alleviate errors in manually annotated data and improve target detection accuracy. Based data on offshore wind turbine blade flaws are used to train the model. An offshore wind turbine blade machine vision identification system incorporates the learned model. The experimental findings demonstrate that the YOLOv5x algorithm, introducing the CBAM attention mechanism compared to the original one, improves the mAP value by 4.71% , the Precision value by 7.48% and satisfies the real-time criteria.

查看全文查看/发表评论下载PDF阅读器