| As a populous country,China’s demand for energy is very huge,among which the demand for electric energy is second to none.With the rapid development of China’s economy after the reform and opening up,the scale of China’s power grid is also expanding.Transmission line as an important carrier of electric energy transmission,its safe and stable operation is an important guarantee of electric energy supply.In reality,foreign objects on transmission lines often cause short circuit and even fire accidents,so it is very necessary to carry out transmission line inspection.In order to reduce the inspection burden of power workers,this thesis studies the foreign objects detection algorithm of transmission lines,introduces YOLOX-s network model,and uses the algorithm to design the detection software.The main work of this thesis is as follows:(1)The Data Augmentation method is used to expand the limited initial datasets of foreign objects,and various image processing methods are used to increase the complexity of the datasets,which lays a foundation for the subsequent experiments.(2)The foreign objects detection algorithm of transmission line is designed by using YOLOX-s network model.Firstly,the structure of each part of YOLOX-s is analyzed in detail.Then,under the same network training strategy,the detection effects of YOLOX-s network model and other deep learning algorithms are compared.The confidence loss function of YOLOX-s was improved to Focal loss function,and ASFF mechanism was adopted.Experimental results show that the YOLOX-s and the improved YOLOX-s have good comprehensive detection effect and can meet the requirements of accuracy and real-time.(3)The transmission line foreign objects detection software is designed,which consists of user login,model update,foreign objects detection and historical data module.The user login is convenient for the user to manage the software,the model update is convenient for the subsequent update of software functions,foreign objects detection as the core content of the software can complete the pictures,videos and cameras for detection,and the historical data can be viewed and cleared.After testing,the content of each module of the software has been realized. |