YOLOV4-Based Wind Turbine Blade Crack Defect Detection

Abstract

Wind turbine blade is an important component of wind turbine. Wind turbine blade crack damage will cause hidden danger to the operation of wind turbine. The current wind turbine blade defect detection mainly relies on manual inspection, and the image detection technology can improve the inspection efficiency and reduce the unit maintenance cost. In view of the existing wind turbine blade crack defect detection algorithm with low recognition rate and low accuracy, a YOLOv4-based wind turbine blade crack detection method is proposed. First establish the wind turbine blade crack image dataset, then the anchor box parameters in YOLOV4 are optimized by K-means++ algorithm to make the anchor box parameters match the crack defect size; BiFPN is used instead of PANet to achieve better feature fusion, and finally the Focal Loss function is introduced to balance the number of small size defect samples in the data. The comparison tests show that the AP of the improved YOLOv4 algorithm reaches 93.49, which is better than the original YOLOv4 and the other three comparison algorithms, and has better efficiency and practicability.