Generative Adversarial Networks based image generation of insulator

In recent years, the combination of artificial intelligence and power systems has been increasing. However, the collection of image datasets of power equipment is limited by places and environments, so the number of collected datasets is relatively small, which makes it unable to provide sufficient data support for specific applications. We propose to use Generative Adversarial Networks to generate images of electrical equipment from existing sparse datasets, thereby increasing the size of the electrical equipment dataset. In this paper, three generative adversarial networks with different structures are used to generate images of insulator and to assess the caliber of images produced by the three models, the analysis of loss function image and FID score of generated images is performed.

I. Introduction

Artificial intelligence is increasingly integrated with the field of electrical engineering. The application of artificial intelligence technology can not only reduce human and material resources, but also improve the accuracy rate. The implementation of these techniques often requires power equipment related datasets as support. However, there are two difficulties in the collection of electric equipment pictures. The first difficulty is that the size of the pictures is not uniform. Because when taking pictures of equipment, the staff uses different devices, such as mobile phones, cameras, drones, etc., so the collected photos are of different sizes, which causes certain troubles when using the dataset. The second difficulty is that the collection of pictures is limited by the site and environment. Most electrical equipment has specific workplaces, which are dangerous and confidential, so no one except professionals can enter, and the collected pictures are not public, causing difficulties in data collection. Therefore, considering the difficulties encountered in the collection of electrical equipment, we believe that it can be solved by combining the techniques of image generation.