Enhancing recognition performance of vortex arrays through conditional generative adversarial network-based data augmentation

Orbital angular momentum (OAM) holds significant potential for achieving extremely high communication capacity, attributed to its orthogonality and infinite modes. Employing convolutional neural networks (CNN) for OAM mode recognition is an effective strategy to mitigate the effects of turbulence. However, recognition accuracy can be compromised when the training dataset is limited. To address this, we leveraged a conditional generative adversarial network (cGAN) for data augmentation (DA). The well-trained cGAN generated abundant augmented data with mode information, thereby enhancing the performance of the CNN. Experimental results clearly demonstrate that cGAN-based DA is an effective method for boosting recognition accuracy, resulting in a significant increase in recognition accuracy, rising from 24% to more than 99%. In addition, analyzing the relationship between the degree of DA and accuracy was instrumental in finding a balance between generation time cost and accuracy improvement. In addition, the application of cGAN-based DA to decomposed OAMs from the vortex array further validates its applicability in enhancing recognition performance.