The analysis of infrared spectroscopy of substances is a non-invasive measurement technique that can be used in analytics. Although the main objective of this study is to provide a review of machine learning (ML) algorithms that have been reported for analyzing near-infrared (NIR) spectroscopy from traditional machine learning methods to deep network architectures, we also provide different NIR measurement modes, instruments, signal preprocessing methods, etc. Firstly, four different measurement modes available in NIR are reviewed, different types of NIR instruments are compared, and a summary of NIR data analysis methods is provided. Secondly, the public NIR spectroscopy datasets are briefly discussed, with links provided. Thirdly, the widely used data preprocessing and feature selection algorithms that have been reported for NIR spectroscopy are presented. Then, the majority of the traditional machine learning methods and deep network architectures that are commonly employed are covered. Finally, we conclude that developing the integration of a variety of machine learning algorithms in an efficient and lightweight manner is a significant future research direction.

Convolutional neural networks (CNNs) for hyperspectral image (HSI) classification have generated good progress. Meanwhile, graph convolutional networks (GCNs) have also attracted considerable attention by using unlabeled data, broadly and explicitly exploiting correlations between adjacent parcels. However, the CNN with a fixed square convolution kernel is not flexible enough to deal with irregular patterns, while the GCN using the superpixel to reduce the number of nodes will lose the pixel-level features, and the features from the two networks are always partial. In this paper, to make good use of the advantages of CNN and GCN, we propose a novel multiple feature fusion model termed attention multi-hop graph and multi-scale convolutional fusion network (AMGCFN), which includes two sub-networks of multi-scale fully CNN and multi-hop GCN to extract the multi-level information of HSI. Specifically, the multi-scale fully CNN aims to comprehensively capture pixel-level features with different kernel sizes, and a multi-head attention fusion module is used to fuse the multi-scale pixel-level features. The multi-hop GCN systematically aggregates the multi-hop contextual information by applying multi-hop graphs on different layers to transform the relationships between nodes, and a multi-head attention fusion module is adopted to combine the multi-hop features. Finally, we design a cross attention fusion module to adaptively fuse the features of two sub-networks. AMGCFN makes full use of multi-scale convolution and multi-hop graph features, which is conducive to the learning of multi-level contextual semantic features. Experimental results on three benchmark HSI datasets show that AMGCFN has better performance than a few state-of-the-art methods.

Single chip integrated spectrometers are critical to bring chemical and biological sensing, spectroscopy, and spectral imaging into robust, compact and cost-effective devices. Existing on-chip spectrometer approaches fail to realize both high resolution and broad band. Here we demonstrate a microring resonator-assisted Fourier-transform (RAFT) spectrometer, which is realized using a tunable Mach-Zehnder interferometer (MZI) cascaded with a tunable microring resonator (MRR) to enhance the resolution, integrated with a photodetector onto a single chip. The MRR boosts the resolution to 0.47 nm, far beyond the Rayleigh criterion of the tunable MZI-based Fourier-transform spectrometer. A single channel achieves large bandwidth of ~ 90 nm with low power consumption (35 mW for MRR and 1.8 W for MZI) at the expense of degraded signal-to-noise ratio due to time-multiplexing. Integrating a RAFT element array is envisaged to dramatically extend the bandwidth for spectral analytical applications such as chemical and biological sensing, spectroscopy, image spectrometry, etc.

Graph can achieve good performance to extract the low-dimensional features of hyperspectral image (HSI). However, the present graph-based methods just consider the individual information of each sample in a certain characteristic, which is very difficult to represent the intrinsic properties of HSI for the complex imaging condition. To better represent the low-dimensional features of HSI, we propose a multistructure unified discriminative embedding (MUDE) method, which considers the neighborhood, tangential, and statistical properties of each sample in HSI to achieve the complementarity of different characteristics. In MUDE, we design the intraclass and interclass neighborhood structure graphs with the local reconstruction structure of each sample; meanwhile, we also utilize the adaptive tangential affine combination structure to construct the intraclass and interclass tangential structure graphs. To further enhance the discriminating performance between different classes, we consider the influence of the statistical distribution difference for each sample to develop an interclass Gaussian weighted scatter model. Then, an embedding objective function is constructed to enhance the intraclass compactness and the interclass separability and obtain more discriminative features for HSI classification. Experiments on three real HSI datasets show that the proposed method can make full use of the structure information of each sample in different characteristics to achieve the complementarity of different features and improve the classification performance of HSI compared with the state-of-the-art methods.

Introduction: A key aim of triage is to identify those with high risk of cardiac arrest, as they require intensive monitoring, resuscitation facilities, and early intervention. We aim to validate a novel machine learning (ML) score incorporating heart rate variability (HRV) for triage of critically ill patients presenting to the emergency department by comparing the area under the curve, sensitivity and specificity with the modified early warning score (MEWS). Methods: We conducted a prospective observational study of critically ill patients (Patient Acuity Category Scale 1 and 2) in an emergency department of a tertiary hospital. At presentation, HRV parameters generated from a 5-minute electrocardiogram recording are incorporated with age and vital signs to generate the ML score for each patient. The patients are then followed up for outcomes of cardiac arrest or death. Results: From June 2006 to June 2008 we enrolled 925 patients. The area under the receiver operating characteristic curve (AUROC) for ML scores in predicting cardiac arrest within 72 hours is 0.781, compared with 0.680 for MEWS (difference in AUROC: 0.101, 95% confidence interval: 0.006 to 0.197). As for in-hospital death, the area under the curve for ML score is 0.741, compared with 0.693 for MEWS (difference in AUROC: 0.048, 95% confidence interval: -0.023 to 0.119). A cutoff ML score ≥ 60 predicted cardiac arrest with a sensitivity of 84.1%, specificity of 72.3% and negative predictive value of 98.8%. A cutoff MEWS ≥ 3 predicted cardiac arrest with a sensitivity of 74.4%, specificity of 54.2% and negative predictive value of 97.8%. Conclusion: We found ML scores to be more accurate than the MEWS in predicting cardiac arrest within 72 hours. There is potential to develop bedside devices for risk stratification based on cardiac arrest prediction.