Improved YOLOv8n Model Applied in Defect Detection for Transmission Line Insulators

To guarantee the stability of power transmission lines and improve the detection accuracy of anomalies in insulators in transmission lines, a detection of defects in insulators algorithm based on improved YOLOv8n is proposed. Firstly, a new robust feature downsampling (RFD) module is implemented to substitute the convolutional downsampling module in the original model of YOLOv8 for the difficulty of identifying broken and flashover defect targets in insulators. Then, a multi-scale feature fusion and diffusion pyramid network (MFDPN) framework is designed in the neck network, which allows each scale feature to have detailed contextual information through the designed multi-scale feature fusion (MSFF) module and diffusion mechanism, which aids in the improved detection of later targets. To further improve target feature extraction and reduce model parameters and computational load, the C2f module in YOLOv8 is upgraded with deformable convolution v3 (DCNv3), forming the C2f-DCNv3 module. This enhanced module is also applied in the necking network, replacing the original C2f module. Results from experiments on the insulator defect dataset demonstrate that the upgraded model effectively decreases parameter numbers, but also achieves a precision of 95% and a recall of 88%, [email protected] of 92.9%, [email protected]:0.95 of 62.1%, which are 3.3%, 2.2%, 3.8%, and 4%, respectively, higher than the original model. It is also compared with several mainstream target detection models, outperforming other models in detection effectiveness.