Multiclass Dataset for Intelligent Detection of Wind Turbine Blade Defects Using Drone Imagery

Achieving intelligent and automated detection of defects in wind turbine blades has become a critical task for contemporary wind farm inspection operations. However, existing datasets for blade defect detection exhibit notable shortcomings, including insufficient defect attributes and limited scale, which hinder the advancement of related detection algorithms. This paper presents a standardized multiclass dataset of visible images of wind turbine blade defects for visual inspection, comprising six categories and 1,065 real blade images captured by unmanned aerial vehicles (UAVs). To provide a comprehensive characterization of this dataset, we conducted a mathematical analysis to identify unique attributes of the defective targets. Additionally, experiments employing manual feature descriptions and six deep learning methods were performed under varying mean average precision thresholds to facilitate both qualitative and quantitative comparisons of the dataset. The proposed dataset and associated model serve as a benchmark for the visual inspection of surface defects on wind turbine blades, thereby promoting research in high-performance defect detection.