YOLO-WTB: Improved YOLOv12n Model for Detecting Small Damage of Wind Turbine Blades From Aerial Imagery
Wind energy has been extensively studied worldwide to advance technology, reduce operating costs, and improve performance. A key challenge in this field is ensuring the optimal performance of wind turbines through proactive and effective maintenance strategies. In particular, wind turbine blade insp...
Saved in:
| Main Authors: | , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
IEEE
2025-01-01
|
| Series: | IEEE Access |
| Subjects: | |
| Online Access: | https://ieeexplore.ieee.org/document/11080388/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Wind energy has been extensively studied worldwide to advance technology, reduce operating costs, and improve performance. A key challenge in this field is ensuring the optimal performance of wind turbines through proactive and effective maintenance strategies. In particular, wind turbine blade inspection and fault detection play an important role in minimizing the risk of unexpected failures, downtime, and operational disruptions. Although predictive maintenance methods based on machine learning, deep learning, and traditional visual inspection have been widely studied, detecting small faults from aerial images remains a major challenge. The main obstacles include data shortages, high computational complexity, limited labelled datasets, and difficulty in accurately identifying faults under real-world conditions. Notably, one of the most pressing problems that modern deep learning models face today is the detection of small-sized objects in images. To address these challenges, we propose an improved model based on the You Only Look Once version 12n model, which enhances the accuracy of wind turbine blade surface damage detection while maintaining real-time processing capability. The improvements are made by adding a very small target Head and removing the two Heads for medium and large targets. In addition, in the backbone part, we also propose to remove a Convolution module and an Area Attention Concatenate-Convolution-Fusion module and add an improved SoftPool Feature Spatial Pyramid Pooling - Fast module to increase the feature extraction ability while maintaining the complexity of the model. The proposed model not only optimizes wind turbine maintenance efficiency but also contributes to advancements in the field of computer vision. |
|---|---|
| ISSN: | 2169-3536 |