Research On Defect Detection Of Transmission Line Based On Deep Learning

The safe and stable operation of transmission lines is an important guarantee for the supply of electricity.However,they are often exposed to harsh natural environments,making them susceptible to defects such as foreign object attachment,insulator damage,and flashover discharge,which seriously affect the stability and safety of the power system.As the scale of China’s power grid continues to expand,traditional manual and drone inspections can no longer meet the daily inspection needs of transmission lines and are gradually being replaced by new inspection methods.Existing transmission line defect detection algorithms cannot meet the real-time and accuracy requirements of transmission line inspections.Therefore,thesis focuses on the detection of foreign objects and insulator defects in transmission lines and conducts research on defect detection using deep learning algorithms.The specific work is as follows:Aiming at the problem of unclear edge features and complex appearance of foreign objects on power transmission lines,an improved YOLOv5 algorithm for detecting foreign objects on transmission lines is proposed.Firstly,the lightweight Ghost bottleneck module is used to replace the standard convolution in the CSP1-x structure with high computational complexity.Additionally,the non-linear activation function in the CA module is replaced with the Swish function to constract an S-CA attention mechanism,which is embedded into the CSP1-x structure to build the lightweight CGA structure.Then,an FE-FPN feature pyramid structure is designed to enable the model to acquire more fine-grained information.Finally,the model detectin accuracy is improved by replacing the more advantageous EIo U Loss.The experimental results show that the improved network model has higher detection accuracy and faster detection speed,making it more suitable for deployment on mobile devices.Aiming at the problem of irregular shape,occlusion from different angles and large morphological differences of insulator defects,a new VSPENet lightweight network structure is designed as the encoding network of CenterNet.Firstly,the Vo VNet network is introduced and a network architecture based on deformable convolution is designed on it.Then,a feature fusion structure is designed using bilinear interpolation to effectively fuse the obtained multi-level feature maps.Finally,the SE attention mechanism and the ASPP module are combined to construct the ASPE structure to capture the feature information of insulator defects at multiple scales.Through comparative analysis and experimental verification,the improved model balances between the detection speed and accuracy,and improves the detection performance of the model.The improved detection models are deployed on the NVIDIA Jetson Nano development board.The results show that the improved lightweight models have good detection performance,with an average frame rate is kept at about 35f/s,which meets the requirements of transmission line defect detection in practical scenarios.