Estimation of Centroid Position of Silicon Particles Based on GAN and Visual Tracking

In this paper, we used generative adversarial network (GAN) and visual tracking to detect the growth of silicon particles. To solve the problem of fewer data sets, we use the Wasserstein-GAN (WGAN) model to expand the data set. From the loss functions of the generator and discriminator, the quality of the data generated by the model is high. The position of the center of mass of the silica particles during the melting process was determined by the extracted tracking mark results, and the motion trajectory of the center of mass of the silica was given.