LG vortex-beam speckle-based underwater surface roughness measurement via a lightweight classification-regression network.

Underwater surface roughness affects flow friction as well as acoustic and optical scattering, making its accurate assessment crucial for the reliable operation of underwater equipment. Since speckle formation depends on both surface microstructure and incident beam structure, Laguerre-Gaussian (LG) vortex beams are introduced as OAM-dependent structured illumination, whose helical phase can be perturbed by rough-surface scattering and manifested as variations in speckle statistics and texture. In this paper, LECR-Net is proposed as a lightweight EfficientNetV2-based classification-guided regression network for underwater surface roughness measurement using LG vortex-beam speckle images. The network compresses the EfficientNetV2 backbone, introduces a stage-wise attention configuration, and uses a conditional residual regression strategy to refine the reference roughness value associated with the predicted class. An underwater LG vortex-beam speckle dataset is experimentally constructed by illuminating steel samples with different machining methods and reference roughness values. Experimental results show that LECR-Net achieves 99.98% accuracy in machining-method and roughness-category classification, keeps the maximum relative error of roughness prediction below 4%, and reduces the model size to 0.86 MB with 0.21 M parameters. These results indicate that the proposed method provides an accurate and lightweight solution for underwater surface roughness measurement based on LG vortex-beam speckle imaging.