Event-based Shack–Hartmann wavefront sensing using spiking neural networks

Abstract

Event-based sensors offer high temporal resolution, high dynamic range and low latency, making them wellsuited for wavefront sensing under high-speed turbulence. Recent work has demonstrated the effectiveness of event-based Shack–Hartmann wavefront sensing; however, many current methods rely on conventional framebased computer vision techniques (e.g., convolutional neural networks) that cannot take full advantage of the asynchronous event stream. By contrast, spiking neural networks are ideally suited for processing event-based sensor data since both operate on discrete, temporally encoded data. In this work, we introduce a spiking neural network framework for wavefront reconstruction from event-based Shack–Hartmann data. At the cost of a slight reduction in reconstruction accuracy, our SNN offers a significantly reduced algorithmic computational cost compared with a CNN, and offers a roadmap towards improved neuromorphic adaptive optics.