A hybrid framework for terahertz imaging through deep learning using a single-point scanning detector

Despite recent progress in non-destructive testing, an efficient, high-precision, low-cost imaging technique remains challenging, terahertz (THz) computational imaging holds great potentials and prospects for achievement of next-generation of non-destructive defect testing in a variety of THz applications. Here, we present a novel hybrid framework to efficiently realize THz computational imaging for defect detection in flexible antennas within the composite sandwich of UAVs, which is based on deep learning algorithms using a single-point scanning detector. By integrating super-resolution convolutional neural network (SRCNN) with edge detection, our method breaks the diffraction limit to reconstruct high-resolution THz images from low-resolution THz images, enabling rapid and accurate defect detection without costly focal plane arrays. The reconstructed high-resolution image has a structural similarity index of around 0.96 and a peak signal-to-noise ratio of about 42 dB, greatly improving imaging quality. In comparison with traditional single-pixel imaging approaches, the proposed method can significantly increase the imaging efficiency. Our concept can offer a new and promising technological path for the non-destructive evaluation of composite structures in aerospace. Additionally, it may further facilitate the engineering application of THz imaging in fields like biomedical detection and electronic device packaging.