Data Expansion for Foreign Object Detection in Power Grid

There have been some researches on the use of deep learning algorithm in the work of power patrol inspection. With some current object detection algorithms as the core, a power grid image recognition system can be built to detect abnormal operation or foreign invasion in the power grid, which could save manpower and material resources and improve the security of the power grid. Deep learning requires a number of effective samples to work. There are few real images of hidden dangers, which cannot meet the demand. This paper aims to explore a feasible data expansion scheme. One possible way is to merge the target with the context background image according to some rules. Another method is to generate samples via GAN (Generative adversarial network). Experiments results show that the performance of the power grid image recognition system is improved by using the extended training set.