Diabetes is considered one of the life-threatening diseases in the world, which needs regular monitoring of blood glucose levels. In this paper, we developed a portable system that makes near-infrared spectroscopy (NIRS) technology available to non-professionals through a mobile application and a specially-made enclosure. It overcomes the shortcomings of traditional spectroscopy systems, such as large volume, high cost, complicated operation, and difficulty in online detection. To verify the feasibility of NIRS in noninvasive blood glucose concentration detection, after the pretreatment of the acquired original spectra, we compared two different feature extraction algorithms of synergy interval (Si) and genetic algorithm (GA). On this basis, two quantitative prediction models of partial least squares (PLS) and extreme learning machine (ELM) were established. The experimental results showed the model based on the combination of Si and GA and ELM (i.e., Si-GA-ELM model) as the most accurate among the selected models. At the same time, the prediction accuracy of the spectral waveband was higher than that of the full. To further overcome the difficulty of establishing a finite sample data model and reduce the influence of individual differences, the model transfer method TrAdaBoost was used to enhance the accuracy and stability of our model. The final experimental results show that the NIR spectrometer used is portable and light and can be encased as a handheld device form. Computation models combining machine learning and chemometric methods make the estimated blood glucose more feasible, which is an innovative work in noninvasive blood glucose measurement fields.

The multi-sensor system based on single pixel imaging can realize the simultaneous measurement and imaging of various optical information (such as spectrum, polarization, etc.), in which a single pixel sensor can measure one kind of optical information. Spectral imaging technology can simultaneously obtain rich information such as spatial structure and fine spectral characteristics of the target scene. The architecture and implementation of most spectral imaging systems are complex, which limits or affects their indicators, parameters, and applications to varying degrees. Here, we propose a novel multispectral single pixel imaging method based on spatial and spectral dimension decoupling. This method uses spatial dimension compression coding modulation, and spectral dimension modulation through ingenious optical system design. It utilizes a single pixel detector array combined with traditional spectral filtering technology to directly and parallel collect spatial image compression coding information from various spectral channels. This method reduces the information carrying capacity of a single photon and reduces the requirements for detection signal to noise ratio. The experiment demonstrated 15 channels of visible and near-infrared spectral single pixel imaging and demonstrated real-time multispectral single pixel imaging with an imaging speed of 5 Hz, which can be extended to spectral imaging with higher spectral resolution and more spectral channels. This work can facilitate the use of multi-sensor arrays for more convenient research such as spectral information acquisition.

With the increase in the number of types of spectrometers in use, calibration models cannot be shared among different instruments; however, this problem can be solved via calibration transfer (CT). In this study, a variety of modern process analysis technology (PAT) data are taken as the research object. After preprocessing the spectra data using principal component analysis (PCA) and cubic spline interpolation, the TrAdaBoost algorithm in transfer learning combined with extreme learning machine (ELM), i.e., TrAdaBoost-ELM, is used to transfer the master model to slave instruments and to make comparisons with the transfer via an extreme learning machine auto-encoder method (TEAM) and the semisupervised parameter-free framework for calibration enhancement (SS-PFCE) method. After the master model is transferred by the TrAdaBoost-ELM algorithm for the prediction dataset of slave instruments, the mean coefficient of determination of prediction ( $R_{\mathrm {p}}^{2}$ ) increases from 0.7843 to 0.8707, and the mean root-mean-square error of prediction (RMSEP) decreases from 2.7508 to 2.3112. Furthermore, variable combination population analysis (VCPA) in combination with a genetic algorithm (VCPA-IGA) was used to select characteristic wavelengths in molecular and atomic spectra, respectively. For the same type of laser-induced breakdown spectroscopy (LIBS) instruments K1 and K2, after processing by the VCPA-IGA algorithm, the LIBS calibration model established on K1 was transferred successfully to K2, and for the major elements, the mean $R_{\mathrm {p}}^{2}$ = 0.9563 and the mean RMSEP = 1.3796. After processing by the VCPA algorithm, the near-infrared (NIR) model for instrument L was transferred to a different instrument J, and the prediction results were $R_{\mathrm {p}}^{2}$ = 0.9110 and RMSEP = 0.4044 °Brix. The results demonstrated that an appropriate variable selection method combined with the TrAdaBoost-ELM algorithm can be effectively used for CT for spectrometers of the same and different types, thus achieving model sharing between different spectrometers.

The ability of underwater robots to accurately perceive their surroundings relies heavily on high-quality imaging systems. However, capturing clear images in aquatic environments is difficult due to light absorption and scattering challenges. Numerous studies have been conducted to develop underwater image enhancement techniques to address this issue, but striking a balance between computational speed, enhancement effect, and robustness remains a significant challenge. Our research takes a unique approach by analyzing the degradation of standard colors and utilizing the degradation of white as a priori information for our proposed adaptive color restoration and histogram equalization method. By modeling the difference in white color between air and underwater images, we estimate compensation coefficients via optimization to restore the color of underwater images. Our method achieves a superior balance of computational speed, color enhancement effect, and robustness compared with other state-of-the-art methods, as demonstrated by our experiments in various sea areas. This research significantly advances our understanding of underwater imaging and provides a practical solution for enhancing underwater images.

The abundant movement information carried by the surface electromyography (sEMG) signal can comprehensively reflect the muscle movement patterns on the surface of the human body, and the characteristics of physiological signals involving movement information have promoted the vigorous development in the field of human-computer interaction in machine learning. This paper takes subtle gestures as the research objects, and proposes a subtle gestures recognition system that uses two-channel sEMG signal and utilizes Quadratic Discriminant Analysis (QDA) and Linear Discriminant Analysis (LDA) which are parameter estimation classifiers based on probability density. First, combining the raw sEMG signal and the envelope sEMG signal, the number of channels for sensors and corresponding position of the muscle groups are preferably selected. Then, we propose to use the Pearson correlation coefficient to optimize the four types of features extracted in the time domain, frequency domain, time-frequency domain and AR model parameters, which effectively ensures the uniqueness of each gesture and eliminates individual differences. Finally, this paper compares four classifiers of QDA, LDA, Support Vector Machine (SVM), and K-Nearest Neighbor (KNN), and explores the pattern recognition classifier suitable for subtle actions. The experiment collects 9 kinds of subtle gestures from 8 subjects. The results show that the system can use only two channels of sEMG signal to recognize 9 kinds of subtle gestures which include multi-finger clicking or pinching or holding objects, and the most suitable classifiers QDA and LDA have the average recognition rates of 95.79% and 95.01%, respectively.