Image Enhanced Mask R-CNN: A Deep Learning Pipeline with New Evaluation Measures for Wind Turbine Blade Defect...

Area of Science:

• Engineering
• Computer Science
• Renewable Energy

Background:

• Growing demand for wind power necessitates efficient wind turbine blade (WTB) inspections.
• Current inspection methods require improvement for defect detection and classification.

Purpose of the Study:

• To evaluate deep learning algorithms (YOLOv3, YOLOv4, Mask R-CNN) for WTB defect detection.
• To propose novel performance metrics for defect detection tasks.
• To develop an optimized defect detection pipeline for WTB.

Main Methods:

• Empirical investigation of YOLOv3, YOLOv4, and Mask R-CNN on a WTB inspection dataset.
• Development of new metrics: Prediction Box Accuracy, Recognition Rate, and False Label Rate.
• Dataset augmentation using rotation and flipping techniques.

Main Results:

• Mask R-CNN demonstrated superior performance across proposed metrics, especially with transformation-based augmentations.
• The best performing dataset yielded mWA values: Mask R-CNN (86.74%), YOLOv4 (78.28%), YOLOv3 (70.08%).
• A new pipeline, IE Mask R-CNN, was proposed, integrating enhanced image processing and a tuned Mask R-CNN model.

Conclusions:

• Mask R-CNN is highly effective for WTB defect detection and classification.
• Image enhancement and augmentation are crucial for optimizing deep learning model performance.
• The proposed IE Mask R-CNN pipeline offers a robust solution for WTB defect analysis.