Terahertz (THz) nondestructive testing (NDT) technology has been increasingly applied to the internal defect detection of composite materials. However, the THz image is affected by background noise and power limitation, leading to poor THz image quality. The recognition rate based on traditional machine vision algorithms is not high. The above methods are usually unable to determine surface defects in a timely and accurate manner. In this paper, we propose a method to detect the internal defects of composite materials by using terahertz images based on a faster region-convolutional neural networks (faster R-CNNs) algorithm. Terahertz images showing internal defects in composite materials are first acquired by a terahertz time-domain spectroscopy system. Then the terahertz images are filtered, the blurred images are removed, and the remaining images are enhanced with data and annotated with image defects to create a dataset consistent with the internal defects of the material. On the basis of the above work, an improved faster R-CNN algorithm is proposed in this paper. The network can detect various defects in THz images by changing the backbone network, optimising the training parameters, and improving the prior box algorithm to improve the detection accuracy and efficiency of the network. By taking the commonly used composite sandwich structure as a representative, a sample with typical defects is designed, and the image data are obtained through the test. Comparing the proposed method with other existing network methods, the former proves to have the advantages of a short training time and high detection accuracy. The results show that the mean average precision (mAP) without data enhancement reached 95.50%, and the mAP with data enhancement reached 98.35% and exceeded the error rate of human eye detection (5%). Compared with the original faster R-CNN algorithm of 84.39% and 85.12%, the improvement is 11.11% and 10.23%, respectively, which demonstrates superb feature extraction capability and reduces the occurrence of network errors and omissions.

Ceramic adhesive structures have been increasingly used in aerospace applications. However, the peaks of the signal on the upper and lower surface of the adhesive layer are difficult to measure directly due to the thin thickness of the adhesive layer and the effect of the attenuation dispersion of the ceramic layer. Thus, the existing non-destructive testing techniques have been ineffective in detecting adhesive quality. In this paper, the thickness of the adhesive layer is measured using terahertz time-domain spectroscopy. A sparse deconvolution method is proposed for the terahertz time-domain spectral signal of ceramic adhesive structures with different adhesive layer thicknesses. The results show that the methods proposed in this paper can realize the separation of reflection signals for glue layers with a thickness of 0.20 mm. By comparing with a wavelet denoising method and a modified covariance method (AR/MCM), the effectiveness of the sparse deconvolution method in estimating the thickness of the glue layer is demonstrated. This work will provide the theoretical and experimental basis for using terahertz time-domain spectroscopy to detect the homogeneity of ceramic adhesive structures.

Multilayer composite materials often develop internal defects at varying depths due to manufacturing and environmental factors. Traditional planar scanning methods lack the ability to pinpoint defect locations in depth. This study proposes a terahertz time-domain spectroscopy (THz-TDS)-based defect detection method using continuous wavelet transform (CWT) to convert spectral signals into time-frequency images. These are analyzed by the ResNet18 model combined with a support vector machine (SVM) classifier. Comparative experiments with four classical deep learning models and three classifiers show that the Residual Network with 18 layers (ResNet18-SVM) approach achieves the highest accuracy of 98.56%, effectively identifying three types of defects. The results demonstrate the method’s strong feature extraction, depth resolution, and its potential for nondestructive evaluation of multilayer structures.

In the case of multiple degradations, current deep-learning-based gray image super-resolution (SR) methods equally process all components in an image, resulting in missing subtle details. To address this issue, we elaborate a cartoon-texture decomposition-based (CTD) module that can automatically decompose an image into a smooth cartoon component and an oscillatory texture component. The CTD module is a plug-and-play prior module that can be applied in solving imaging inverse problems. Specifically, for the SR task under multiple degradations, we apply CTD as a prior module to build an unfolding SR network termed CTDNet. For the SR task of real terahertz images, the boundary (i.e., the boundary between the object of interest and the carrier table) recovered by CTDNet has artifacts, which limits its realistic applications. To reduce these boundary artifacts, we post-process the SR terahertz images by using a boundary artifact reduction method. Experimental results on the synthetic dataset and real terahertz images demonstrate that the proposed algorithms can maintain subtle details and achieve comparable visual results. The code can be found at https://github.com/shibaoshun/CTDNet.

The asymmetric transmission of terahertz (THz) waves has attracted considerable interest because of its novel applications in isolators and wireless communication. An A-shaped switchable asymmetric transmission metasurface integrating vanadium dioxide (VO₂) in the THz band is proposed, which achieved dynamic control of asymmetric transmission by exploiting insulator-to-metal transition characteristics of VO₂. When VO₂ is in an insulating state, the proposed metadevice realized an asymmetric transmission effect of the asymmetric transmission parameter reaching up to 0.76 and realized linear polarization conversion with a 95% polarization conversion rate. Furthermore, the proposed metadevice prevented asymmetric transmission and polarization conversion as VO₂ transforms in the metallic state. This work provides a new ideal for the design of switchable asymmetric transmission metadevices and has potential applications in wireless communication, sensing, and polarization control.

Baicalin concentration is concerned in manufacture of scutellaria spray drying powder as a traditional Chinese medicine, and the quality control based on high‐performance liquid chromatography is inconvenience. In this study, terahertz time domain spectroscopy was employed to achieve quality control of scutellaria spray drying powder; however, an acute difficulty was found that terahertz spectra overlapped due to the disturbance in both content matrix and measurement error. In this study, similar terahertz spectra of scutellaria spray drying powder were classified with the help of Gaussian mixture model and built a classifier based on probability feature instead of spectral features conventionally employed in previous investigations. To explore the feasibility of GMM, principal component analysis was given, indicating that it is possible to train GMM with original features and proper principal components. Probable advantage of training GMM based on PCA feature was discussed and so it was with the capacity of the model to identify the linear combined spectra by comparing the performance of GMM and a decision tree model. Above all, the reason why GMM shows potential in the analysis of TCM terahertz spectra was illustrated by comparing the thought of discriminative model and generative model. This study implied that generative model may have natural advantage of overcoming the inherent disturbance of terahertz spectroscopy, which would be promising in future studies.