This research reports the design and analysis of a multiband reflective circularly polarized (CP) converter constructed on a 5 \(\mu m\) thick quartz (SiO2) substrate. The CP converter is optimized by a CNN based architecture which extracts key features from it’s frequency response data using convolutional layers. It enhances learning using residual connections and batch normalization. It reduces overfitting with dropout layers and global average pooling. A graphene strip is positioned at an oblique angle of 45 \(^\circ\) for bringing in tunability in CP generation. The proposed CP converter can be leveraged for THz imaging in detecting kidney stones. The difference in the dielectric permittivity of the kidney with and without stones alter the reflective characteristics of the CP converter. The S parameters are then used to reconstruct images by sum and delay algorithm which can be leveraged for detecting kidney stones.
