Large model object detection technology for defects in transmission line fittings based on transformer

Transmission line fittings, exposed to complex environments for a long time, result in a large number of power fitting images due to the complicated tower-line coupling system. This leads to a heavy inspection workload and poor accuracy. To address this issue, a large model object detection technology for defects in transmission line fittings based on Transformer is proposed. Using a generative adversarial network, a fitting image generation model is built, which generates target images through a generator and a discriminator to enrich the training sample dataset. An evaluation index for credibility is defined for uncertain defect detection results, and an adaptive adjustment mechanism for the encoding layers of the Transformer network is constructed. A Transformer model library with different encoding layers is established to obtain multi-level differentiated features of target images with multimodal defects. Through model training with a large number of samples of transmission line fitting defects, the model automatically learns how to identify and judge fitting defects, achieving object detection for fitting defects. Experimental results demonstrate that this method has a low rejection rate for detecting defects in transmission line fittings and can accurately detect various defects.