We demonstrated a novel fiber in-line Mach-Zehnder interferometer (MZI) with a large fringe visibility of up to 17 dB, which was fabricated by misaligned splicing a short section of thin core fiber between two sections of standard single-mode fiber. Such a MZI could be used to realize simultaneous measurement of tensile strain and temperature. Tensile strain was measured with an ultrahigh sensitivity of −0.023 dB/μɛ via the intensity modulation of interference fringes, and temperature was measured with a high sensitivity of 51 pm/°C via the wavelength modulation of interference fringe. That is, the MZI-based sensor overcomes the cross-sensitivity problem between tensile strain and temperature by means of different demodulation methods. Moreover, this proposed sensor exhibits the advantages of low-cost, extremely simple structure, compact size (only about 10 mm), and good repeatability.

Chest imaging plays an essential role in diagnosing and predicting patients with COVID-19 with evidence of worsening respiratory status. Many deep learning-based approaches for pneumonia recognition have been developed to enable computer-aided diagnosis. However, the long training and inference time makes them inflexible, and the lack of interpretability reduces their credibility in clinical medical practice. This paper aims to develop a pneumonia recognition framework with interpretability, which can understand the complex relationship between lung features and related diseases in chest X-ray (CXR) images to provide high-speed analytics support for medical practice. To reduce the computational complexity to accelerate the recognition process, a novel multi-level self-attention mechanism within Transformer has been proposed to accelerate convergence and emphasize the task-related feature regions. Moreover, a practical CXR image data augmentation has been adopted to address the scarcity of medical image data problems to boost the model's performance. The effectiveness of the proposed method has been demonstrated on the classic COVID-19 recognition task using the widespread pneumonia CXR image dataset. In addition, abundant ablation experiments validate the effectiveness and necessity of all of the components of the proposed method.

The self-organizing fuzzy (SOF) logic classifier is an efficient and non-parametric classifier. Its classification process is divided into an offline training stage, an online training stage, and a testing stage. Representative samples of different categories are obtained through the first two stages, and these representative samples are called prototypes. However, in the testing stage, the classification of testing samples is completely dependent on the prototype with the maximum similarity, without considering the influence of other prototypes on the classification decision of testing samples. Aiming at the testing stage, this paper proposed a new SOF classifier based on the harmonic mean difference (HMDSOF). In the testing stage of HMDSOF, firstly, each prototype was sorted in descending order according to the similarity between each prototype in the same category and the testing sample. Secondly, multiple local mean vectors of the prototypes after sorting were calculated. Finally, the testing sample was classified into the category with the smallest harmonic mean difference. Based on the above new method, in this paper, the multiscale permutation entropy (MPE) was used to extract fault features, linear discriminant analysis (LDA) was used to reduce the dimension of fault features, and the proposed HMDSOF was further used to classify the features. At the end of this paper, the proposed fault diagnosis method was applied to the diagnosis examples of two groups of different rolling bearings. The results verify the superiority and generalization of the proposed fault diagnosis method.

Unsupervised anomaly detection defines an abnormal event as an event that does not conform to expected behavior. In the field of unsupervised anomaly detection, it is a pioneering work that leverages the difference between a future frame predicted by a generative adversarial network and its ground truth to detect an abnormal event. Based on the work, we improve the ability of video prediction framework to detect abnormal events by enhancing the difference between prediction results for normal and abnormal events. We incorporate super-resolution and self-attention mechanism to design a generative adversarial network. We propose an auto-encoder as a generator, which incorporates dense residual networks and self-attention. Moreover, we propose a new discriminator, which introduces self-attention on the basis of a relativistic discriminator. To predict a future frame with higher quality for normal events, we impose a constraint on the motion in video prediction by fusing optical flow and gradient difference between frames. We also introduce a perception constraint in video prediction to enrich the texture details of a frame. The AUC of our method on CUHK Avenue and Shanghai Tech datasets reaches 89.2% and 75.7% respectively, which is better than most existing methods. In addition, we propose a processing flow that can realize real-time anomaly detection in videos. The average running time of our video prediction framework is 37 frames per second. Among all real-time methods for abnormal event detection in videos, our method is competitive with the state-of-the-art methods.

Bulky fiber spectrum analyzers constrain the application of white light interference-based demodulation of Fabry-Perot interferometric (FPI) sensors in practical measurement scenarios. The key to the miniaturization of the fiber Spectrometer is the reconstruction of the fiber spectrum. However, advances in this field are often characterized by a reliance on microelectromechanical systems (MEMS) that cannot have both excellent frequency and resolution. This letter presented a neural network-based systematic scheme to achieve fiber optic sensor’s spectral reconstruction with high-frequency and high-resolution potential. The solution can reconstruct the FPI sensor spectrum consisting of hundreds of points based on extremely sparse (even single-digit) sampling points, with MHz-level frequency, and supports custom tuning the reconstructed wavelength range. This solution can provide a novel platform for designing compact, high-performance fiber spectrum analyzers.

This paper proposes a complete framework for the manufacturing and dual-parameter demodulation of sensor systems consisting of cascaded fiber Bragg gratings (FBGs) and Fabry-Perot interferometers (FPIs). The proposed system aims to enable simultaneous interrogation of the external strain and temperature using a machine learning technique, with limited data availability and without requiring an optical spectrum analyzer. The system converts the sensing signals of a cascaded FBG-FPI sensor into changes in the transmitted optical intensity across multiple channels of the array waveguide grating (AWG). By inputting the transmitted optical intensity and wavelength shift data into a neural network, an intricate nonlinear relationship is established, which enables precise interrogation of the peak wavelength. Furthermore, to overcome the common limitation of insufficient data in data-driven models, we introduce a high-performance data augmentation method that utilizes a generative adversarial network (GAN) to rapidly augment the dataset during the training process. Extensive experiments using a real-world dataset generated in an industrial optical fiber sensing scenario demonstrate the effectiveness and superiority of the proposed framework.