Multi-parameter identification of aberrated vortex beams under slant-path atmospheric turbulence using a multi-task convolutional neural network.

Abstract

Orbital angular momentum (OAM) mode recognition of vortex beams is essential for free-space optical communication; however, it becomes considerably more challenging under slant-path propagation where atmospheric turbulence and optical aberrations jointly distort the beam structure. Here, we propose an improved multi-task residual network, termed DCA-ResNet18, for the simultaneous recognition of the topological charge, spherical aberration coefficient, and coma aberration coefficient of vortex beams under slant-path atmospheric turbulence. The model incorporates dual 3×3 input convolutions, task-adaptive branches, and a convolutional block attention module (CBAM) to enhance the extraction of ring, dark-core, and asymmetric distortion features. Three datasets corresponding to single-coma-aberration, single-spherical-aberration, and dual-aberration condition are generated using a slant-path multi-phase-screen propagation model. The proposed model achieves near-perfect accuracy under single-aberration conditions over different zenith angles (0°, 60°, 80°) and propagation distances (2000, 3000, and 4000 m). Under combined-aberration conditions, the recognition accuracy of all tasks remains above 98% even at a zenith angle of 80° and a propagation distance of 3000 m, demonstrating strong robustness under severe distortion conditions. Compared with ResNet18, ResNet34, Xception, and EfficientNet-B0, DCA-ResNet18 consistently achieves superior performance in dual-aberration recognition. These results demonstrate the effectiveness and robustness of the proposed framework for simultaneous multi-parameter recognition of vortex beams under slant-path atmospheric turbulence with optical aberrations.