Modes recognition in high-power fiber laser by convolutional neural networks

In this paper, mode distribution in large-mode-area (LMA) 25/400 fiber was investigated while attempts to recognize and sort different modes with their combination were carried out on a CNN net via Tensorflow. VGG16 model was chosen as the backbone net through several test to increase precision. The model was trained on a dataset including 6000 pictures in 15 categories. And the final accuracy was up to 0.98. It indicates that recognizing modes in high power fiber laser system based on a CNN net was a feasible plan in the mode control assignment.