CMFCANet: A Wind Turbine Blade Damage Recognition Network Based on Color-Multiscale Dual-Feature Fusion and Channel Attention Mechanism

Drone inspections for detecting wind turbine blade (WTB) defects are efficient but necessitate advanced feature recognition methods for aerial imagery. This article introduces CMFCANet, a novel framework designed to accurately characterize WTB defect features through multifeature fusion. The RGB-HSI color space fusion module facilitates the integration of color information at the feature level, significantly enhancing the extraction of color and shadow details in defect images. Additionally, the proposed network employs a feature pyramid for multiscale feature fusion, thereby improving its adaptability to defects of varying sizes. The channel attention module (CAM) dynamically assigns weights to individual feature channels, emphasizing the most salient features that influence model decisions. Furthermore, class activation maps are utilized to analyze the model’s feature representation, enhancing interpretability. To validate the effectiveness of the proposed method, a comprehensive dataset comprising six distinct types of WTB defects, ranging from minor to severe damage, has been constructed. Experimental results indicate that CMFCANet outperforms existing state-of-the-art (SOTA) models, achieving an overall accuracy of 84.211%. Moreover, the proposed model demonstrates balanced performance, with precision, recall, $F1$ score, and kappa metrics that compare favorably to other methods.