Superposed OAM Unsupervised Recognition in Oceanic Turbulence via Alpha Evolution Optimized Extreme Learning Machine

Orbital angular momentum (OAM) beams based on Laguerre–Gaussian (LG) modes are promising for underwater optical communication, but oceanic turbulence degrades their modal features and recognition performance. Although machine learning is effective for OAM recognition, most methods rely on large labeled datasets, which are costly to obtain underwater. Unsupervised learning alleviates this limitation by enabling label free representation learning, reducing data costs and improving adaptability in complex marine environments. Motivated by these advantages, this paper proposed an Alpha Evolution optimized unsupervised extreme learning machine (AE-US-ELM). Without labeled samples, intensity features of OAM-carrying LG beams under varying turbulence and transmission distances are used for training. In addition, the AE algorithm optimizes US-ELM hyper parameters for different turbulence strengths and transmission distances. The constructed autonomous model enables analytical OAM mode identification without manual parameter tuning, while delivering superior recognition performance. Experimental results show that AE-US-ELM consistently achieves superior performance across turbulence strengths and distances, outperforming several representative unsupervised and optimization-based approaches, providing a computationally efficient and robust solution for practical underwater scenarios.