Research On Foreign Objects Detection Algorithm Of Transmission Lines Based On YOLOv5

As the main carrier of power transmission,transmission lines play a vital role in the safe and stable operation of the power system.In daily life,transmission lines are often damaged by foreign objects,resulting in accidents such as single-phase grounding and interphase short circuit of the ground conductor,and if foreign objects cannot be detected in time,they will even cause casualties and fires.Therefore,regular inspection of transmission lines and timely elimination of foreign objects are important tasks to maintain the safety of the power system.However,traditional manual inspection has problems such as low efficiency,high risk,and weak mobility,in order to achieve fast and accurate detection of foreign objects,in this paper,a foreign objects detection method for transmission lines based on YOLOv5 algorithm is studied.The details of this research work are as follows:(1)Construction of transmission lines foreign objects image dataset.Collect training samples through power patrol images and network public data.Data augmentation was performed on the original dataset to increase the number of image samples and optimize the dataset quality.Manually label the foreign object targets in each image with bounding boxes and labels,and then divide the dataset into training set,verification set,and test set according to a certain proportion to complete the construction of the dataset.(2)A lightweight based YOLOv5 algorithm for transmission lines foreign objects detection(YOLOv5-Lite)is proposed.Using two lightweight improvement strategies to transform its network based on YOLOv5 algorithm.Firstly,the lightweight convolutional neural network Rep VGG is used to lighten the backbone network of YOLOv5,and its network width is adjusted to reduce the network size.Secondly,the lightweight convolution module Ghost module is introduced,and C3 Ghost and GBS modules are designed for the neck network to further compress the network parameters.Experimental results have shown that YOLOv5-Lite can significantly reduce the generation of network parameters while only reducing the detection accuracy by 1.2%.Its model size is compressed to 8.3 MB,about one tenth of YOLOv5.(3)An attention and multilevel feature fusion based YOLOv5-Lite algorithm for transmission lines foreign objects detection(AM-YOLOv5-Lite)is proposed.Aiming at the poor performance of foreign objects detection in special scenarios,the network structure is improved based on YOLOv5-Lite to enhance the detection ability of the algorithm.Considering the weakening and interference of complex background on foreign object feature expression,CBAM attention mechanism is introduced to focus the network on important feature information.Aiming at the problem of weak recognition ability of small target foreign objects,a multi-level feature fusion layer is designed to integrate different receptive field information,and adaptive generation of feature maps at all levels to output weights to dynamically optimize the feature representation of small target foreign objects.Further replace the coupled detection head of the YOLOv5 algorithm with a decoupled detection head to decouple separate feature channels for positioning and classification tasks,improving target positioning capabilities.The experimental results show that the detection accuracy of AM-YOLOv5-Lite has been improved by 1.8% on the basis of YOLOv5,and the detection speed FPS has been increased by about twice.The transmission lines foreign objects detection method proposed in this paper can achieve accurate detection of transmission line foreign objects in complex environments.Its network parameters are low,the model is small,and it has both detection efficiency and accuracy.It achieves timely detection and early handling of suspended foreign matters,reducing economic losses.