改进YOLOv8的风机叶片多尺度缺陷检测

打开文本图片集

关键词：风机叶片；缺陷检测；YOLOv8;多尺度特征；小目标;注意力机制中图分类号：TP391.41;TP183 文献标识码：Adoi：10.37188/OPE.20253309.1496 CSTR：32169.14.OPE.20253309.1496

Improvement of YOLOv8 for multi-scale defect detection in wind turbine blades

ZHU Guang 1，2* ， GU Chen¹，XU Liyun²，SHI Yanqiong¹，DING Zhengyang'， ZHANG Xu1， ZHANG Yonghual

（1. School ofMechanical and Electrical Engineering，Anhui Jianzhu University，Hefei 23O6O1， China; 2.School ofMechanical Engineering， Tongji University，Shanghai 2Ol8O4， China） * Corresponding author，E-mail： guangzhul23@ahjzu. edu. cn

Abstract：Toaddress the challenges of low accuracy，missed detection，and false detection in defect identification of wind turbine blade，an enhanced algorithm based on YOLOv8 is proposed. Initially，a DEC2f module is introduced，replacing the bottleneck structure with a dual convolution kernel design based on efficient multi-scale atention，thereby improving the network's multi-scale feature extraction capability. Subsequently，a global receptive field feature fusion module （GRE-SPPF） is implemented to enhance the capture of global feature information and expand the receptive field.Further improvements include the addition of a small-object detection layer and a multi-scale feature fusion structure in the Neck，optimizing detection performance for small and complex objects. An atention and convolution fusion module （ACFM）is also integrated before the detection head to prioritize critical information while mitigating background interference. Experimental results on a wind turbine blade defectdataset indicate that the proposed algorithm achieves mAP @0.5 and mAP@0.5：0.95 values of 91.1% and 61.8% ，respectively，marking improvements of 6.2% and 6.4% over the baseline algorithm． The recall rate reaches 84.9% ，a 7.7% enhancement，with no substantial increase in computational parameters，demonstrating the algorithm's efficacy for practical wind turbine blade defect detection.

Key words：wind turbine blade；defect detection；YOLOv8；multi-scale features；smal object；attention mechanism

1引言

风力发电技术作为一种清洁且可再生的能源越来越受到关注。（剩余22973字）