Recognition of fractional vortex optical modes with a lightweight convolutional neural network

Abstract Structured light with orbital angular momentum (OAM) provides a theoretically infinite expansion dimension of optical signals, playing a crucial role in optical communications, imaging and information processing, etc. Its applications in these fields require accurate measurement and recognition of the OAM light of fine fractional topological charges. In this work, we develop a fractional OAM light recognition method based on deep learning, achieving accurate recognition of the fine fractional OAM light of a few-shot under low light intensity. We employ a lightweight convolution neural network MobileNetV2 to identify the fractional OAM pictures, which requires few computing resources and can be realized on mobile phones. The performance of our method has been validated by multiple rounds of testing, demonstrating high metrics including accuracy, precision, recall rate, and F1-score. Our work suggests a potential pathway for identifying the OAM light with simple devices and few resources, such as mobile phones and other edge computing devices.