Foreign Object Detection of Transmission Towers Based on Attention Mechanism

We present a transmission tower foreign object detection algorithm that integrates the attention mechanism. Firstly, convolutional neural networks (CNNs) are employed for feature extraction at various scales from transmission tower images. Subsequently, a Transformer-based self-attention mechanism is combined to achieve global and local feature extraction. Finally, the Squeeze-and-Excitation attention mechanism is utilized to optimize the network's focus on specific regions, further enhancing its capability to recognize foreign objects. Experimental results demonstrate that compared with traditional methods, our method can automatically learn and focus on important regions in the image, automatically adjust the weights of features at different scales, and improve the detection performance of small and large scale foreign objects. It effectively detects four types of foreign objects on transmission towers, namely kites, balloons, bird nests, and debris, offering a novel approach to transmission tower safety monitoring.