Pin-missing defect recognition based on feature fusion and spatial attention mechanism

As the critical fasteners on transmission towers, bolts greatly influence transmission lines' safety and operational life. Due to manual inspection's heavy workload and inefficiency, automatic defect detection based on machine learning has gradually become the mainstream in recent years. However, since the bolts occupy a tiny proportion in aerial images and are easily confused with the background, the existing methods cannot satisfy pin-missing detection. Thus, this paper proposes a pin-missing defect detection model based on feature fusion and spatial attention mechanism. On the one hand, a high-resolution feature pooling method using bilinear interpolation is constructed to enhance the representation of small targets. On the other hand, an attention mechanism is designed to capture the global features from different channels and combine their weights to improve classification accuracy. The results show that the average accuracy of the proposed method is 11.63% higher than that of the feature pyramid network. (C) 2021 The Author(s). Published by Elsevier Ltd.