Mode parameter recognition of high-order radial Laguerre-Gaussian beam superposition states using SKCSWT-Net

To efficiently demultiplex the modal parameter information encoded in superposed vortex beams propagating through turbulent atmospheric channels, a model named SKCSWT-Net is proposed. This model integrates the strengths of the Selective Kernel (SK) module and the cross-shaped window transformer (CSWT) to simultaneously identify the azimuthal and radial indices of high-order radially superimposed Laguerre-Gaussian beams (LGBs). Experimental results demonstrate that, under randomly varying atmospheric turbulence of unknown intensity, the model achieves a recognition accuracy of 99.03% for the modal parameters of 16 types of superposed LGBs transmitted over a distance of 1000 meters. Furthermore, on additional test sets with a transmission distance of 2000 meters and a beam wavelength of 850 nm, the model attains recognition accuracies of 89% and 97%, respectively. In robustness evaluations involving occlusion interference and pixel loss, SKCSWT-Net consistently maintains a recognition accuracy exceeding 84.91%.