Research on Transmission Line Hardware Identification Based on Improved YOLOv5 and DeblurGANv2

To precisely identify high voltage transmission line fittings and guide a line patrol robot in executing obstacle-crossing maneuvers tailored to diverse fittings, a real-time detection network based on deep learning DeblurGANv2 and YOLOv5 target detection algorithm was proposed. Addressing challenges inherent in the YOLOv5 algorithm, characterized by a substantial network model, high computational overhead, and diminished operational efficiency, a judicious lightweight refinement is undertaken to enhance the algorithm’s detection speed. The original network was improved by replacing CSPDarknet-53 with Shufflenetv2 and integrating an ECA module into Shufflenetv2, thereby enhancing the model’s focus on relevant features. Simultaneously, the DeblurGANv2 super-resolution reconstruction algorithm is employed to rectify the blurriness of the fittings’ images, resulting in the production of high-quality images and enhancing the accuracy of recognition. Experimental results obtained from self-created datasets demonstrate that the proposed method exhibits a reduction of 50.2% in parameter quantity and a decrease of 62.7% in weight file size when compared to the original YOLOv5 model. Additionally, the proposed method achieves a 1.3% increase in mAP@(0.5). The proposed method demonstrates successful detection of high voltage transmission line fittings, thereby ensuring the accuracy of fuzzy target detection.