Comparative Analysis of Deep Learning Models for Detecting Image Artefacts in Silicon Hyperbolic Metamaterials

Hyperbolic metamaterials (HMMs) have emerged as attractive platforms for deep sub-wavelength imaging across the electromagnetic spectrum due to their distinct hyperbolic dispersion characteristic. However, the actual applicability of HMMs for broad-band imaging is hampered by undesirable imaging artifacts. In this work, we examine the image transmission characteristics of silicon-based HMMs at THz frequencies. Our numerical simulations show that, while the suggested metamaterial can resolve feature sizes below the diffraction limit, it produces significant image artifacts. To overcome this issue, we use a range of deep learning models such as VGG16, MobileNet, SqueezeNet, and AlexNet to categorize visual artifacts. Through considerable testing, we obtain an astonishing classification accuracy of 100%, proving the efficiency of deep learning approaches in reducing image artifacts in HMM-based imaging systems. This research establishes the path for the creation of more durable and dependable imaging systems based on hyperbolic metamaterials.