High-speed quantitative laser speckle computational imaging via event camera and SNN-CNN hybrid architecture
Quantitativi laser speckle contrast imaging (LSCI) is constrained by the frame rate bottlenecks of conventional cameras, making it difficult to satisfy the demands of ultra-high-speed and real-time flow velocity measurements. Event cameras offer a high-dynamic optical perception paradigm with microsecond-level temporal resolution. However, existing spatiotemporal autocorrelation algorithms based on event streams incur massive computational overheads, restricting their advancement towards real-time intraoperative monitoring. This paper proposes a lightweight hybrid Spiking Neural Network and Convolutional Neural Network (SNN-CNN) computational imaging framework. By employing sparse voxelisation, this framework directly extracts high-frequency features from asynchronous event streams, achieving a low-latency end-to-end inference of 432ms. Furthermore, to address the block artefacts in heatmaps at microscopic scales, a dual transposed convolution architecture is introduced for algorithmic compensation, effectively restoring a continuous and smooth flow velocity field that complies with physical laws. Translation experiments using a frosted glass slide demonstrate that the system exhibits high quantitative velocity measurement accuracy and spatial robustness across a wide range of flow velocities, providing a novel computational imaging approach and engineering paradigm for real-time quantitative optical measurement under extreme conditions.