Study On Limiting Factors Of Image Acquisition And Defect Classification Of Photovoltaic Modules Based On Uavs

The role of the photovoltaic industry in future development will become increasingly important.From the current point of view,the growth of photovoltaic power generation is gradually changing from construction-type growth to operation and maintenance-type growth.Aiming at the problem that the photovoltaic power plant inspection and fault diagnosis is subject to complex terrain,the application of the intelligent inspection and fault diagnosis strategy of the photovoltaic power plant combined with the rotor UAVs and computer vision has become a more effective solution at present.This article will focus on the overall solution mentioned above,and provide solutions to several key problems faced by the solution in practical applications:Firstly,this article reviews the current status of domestic and foreign research in image augmentation and task planning,and provides ideas for the subsequent development of the topic.At the same time,the causes and the texture structure of the six types of photovoltaic module visible light defects involved in this paper are introduced.Based on the existing research in the laboratory,a Convolutional Neural Network(CNN)model was built to classify the photovoltaic module defect images.Then aiming at the problem of insufficient training data involved in the training of classification models,this paper separately builds deep learning-based Wasserstein adversarial generation network(WGAN)and cycle adversarial generation network(CycleGAN)and proposes an image augmentation method based on color space analysis and layer overlay to augment the PV module defect images data.The augmented results are combined with the original data to form a new database to train the classification model.Through comparison,explain the advantages and disadvantages of each augmentation method and the adaptation scenario.Finally,in order to simplify the steps of task planning and fault diagnosis,this paper proposes a simulated annealing algorithm suitable for the mobile terminal to optimize the transfer path between task points based on the task point model of the photovoltaic power station.The task planning algorithm and the classification model trained by augmented dataset are transplanted to the mobile terminal,so that the mobile terminal integrates the preliminary overall plan based on the basic functions.