Compressive Sensing Imaging Using Caustic Lens Mask Generated by Periodic Perturbation in a Ripple Tank

TL;DR

This paper presents a novel single-pixel terahertz imaging method using a caustic lens mask generated by a ripple tank, combined with neural network classification, achieving high accuracy with fewer measurements.

Contribution

It introduces a new caustic lens mask technique for compressive sensing in terahertz imaging, reducing measurement time and enabling effective target classification.

Findings

Achieved 95.16% classification accuracy.

Reduced measurement time via dynamic caustic lens sampling.

Demonstrated effective target differentiation with neural networks.

Abstract

Terahertz imaging shows significant potential across diverse fields, yet the cost-effectiveness of multi-pixel imaging equipment remains an obstacle for many researchers. To tackle this issue, the utilization of single-pixel imaging arises as a lower-cost option, however, the data collection process necessary for reconstructing images is time-consuming. Compressive Sensing offers a promising solution by enabling image generation with fewer measurements than required by Nyquist's theorem, yet long processing times remain an issue, especially for large-sized images. Our proposed solution to this issue involves using caustic lens effect induced by perturbations in a ripple tank as a sampling mask. The dynamic characteristics of the ripple tank introduce randomness into the sampling process, thereby reducing measurement time through exploitation of the inherent sparsity of THz band signals. In this study, a Convolutional Neural Network was used to conduct target classification, based on the distinctive signal patterns obtained via the caustic lens mask. The suggested classifier obtained a 95.16 % accuracy rate in differentiating targets resembling Latin letters.