Research On Multi-target Detection Algorithm Of Transmission Line Based On Sample Generation

Ensuring the stable and safe operation of the power system is an important part of the work of various national power grid companies.The power grid company has established a special power inspection department to troubleshoot transmission line components and replace faulty components in a timely manner.At present,UAV patrol line has good prospects as a key research project in the field of electrical engineering.This thesis studies the detection of insulators and tower bird nests that are prone to failure in aerial images,which provides a theoretical basis for the identification of transmission line defects.In deep learning,the richer the sample set,the better the detection effect,but there are not too many hidden danger samples in practice.The experimental data is the national grid data published on Github.Since there are not too many hidden danger samples in practice,a sample was studied.Expanded generation algorithm.Using Poisson fusion to process the foreground and background to generate defective images,and after the opening and closing operations of the generated images are processed to simulate real environment image processing,a transmission line detection database is constructed to improve the generalization performance of the training model.After analyzing the mainstream deep learning-based target detection algorithm,in order to solve the problem of low recognition accuracy of small targets and overlapping targets in the current mainstream deep learning algorithm in object recognition,this thesis proposes a new feature pyramid structure Make full use of the information of the feature map,shorten the fusion path between the lower-level features and the top features,and enhance the role of the underlying features in the entire feature hierarchy.In the algorithm,Res Net50,Res Net101 and Res Net101 + FPN incomplete network proposed in this thesis are used for feature extraction.At the same time,Faster R-CNN and Cascade R-CNN are used as the algorithm framework for data mining of UAV images.Insulators,interphase rods,anti-vibration hammers,bird nests and two types of insulator defects have eight different types of targets for identification tasks.Through a large number of comparative experiments,this method can have good accuracy for small targets and various types of defects in a complex background.The m AP in the 2000 test samples reached 93.5%.The thesis has 35 pictures,9 tables,and 45 references.