ABSTRACT This paper introduces MEA‐YOLO, a robust and efficient detector evolved from YOLOv8n, aimed at achieving an optimal balance between detection accuracy and computational cost in challenging transmission‐line environments: (1) a MAF‐Enchancer module that amalgamates receptive‐field attention convolution (RFA‐Conv), progressive multi‐scale feature refinement and a content‐driven enhancement block to augment feature extraction in instances of substantial degradation; (2) a SENetV2‐based dynamic channel attention technique to accentuate defect‐related representations while mitigating background noise; A dynamic drenching and bifurcation detection mechanism to improve multi‐scale feature modelling while ensuring efficient inference. To assess the suggested methodology, we create a transmission‐line defect dataset encompassing various environmental circumstances and subsequently validate the model on an additional industrial defect dataset to evaluate model generalisability. The experimental findings indicate that MEA‐YOLO attains a mean Average Precision (mAP) of 0.8031 across five demanding scenarios, surpassing the YOLOv8n baseline by 0.0502. Ultimately, thorough comparisons with traditional detectors, current YOLO benchmarks, lightweight alternatives and state‐of‐the‐art transformer‐based real‐time detectors reveal that MEA‐YOLO attains an ideal equilibrium among robustness, accuracy and computational efficiency, validating its practical applicability for real‐world transmission‐line inspection contexts.
Xie et al. (Thu,) studied this question.