Most faults in the hydraulic systems of engineering machinery are related to changes in the composition of hydraulic oil, among which particle contamination is the most common type. To provide spectral dimension with abundant information content of hydraulic oil for hydraulic system monitoring, this study focuses on the characteristics of transmittance spectra of hydraulic oil contaminated by particles. L-HM 68 hydraulic oil samples contaminated by SiO2, Fe, and Cu particles of 25 μm, 10 μm, and 5 μm were prepared, and the particle mass concentration ranged from 0 to 10000 mg/L. Levels of contamination of these oil samples under the ISO 4406 standard were determined by a contaminant particle counter, and their transmittance spectra were obtained by an assembly device in the wavelength range of 200–2500 nm. It was found that the obvious absorption valleys of spectral transmittance curves of these oil samples are mainly distributed in ultraviolet (UV) and near-infrared (NIR) bands. The minimum transmittance values of these absorption valleys are located near seven characteristic wavelengths: 340 nm, 920 nm, 1030 nm, 1200 nm, 1390 nm, 1750 nm, and 2250 nm. Spectral transmittance curves of these oil samples exhibit distinct variation characteristics within each order of magnitude of mass concentration. Smaller contaminant particles lead to a faster drop in the spectral transmittance curve of hydraulic oil with increasing mass concentration. Transmittance at each characteristic wavelength shows exponential relationships with mass concentration. It was determined that the detector operating near 1030 nm serves as a relatively suitable option for the transmittance spectrum detection of hydraulic oil. For the relationship between transmittance at the characteristic wavelengths and the level of contamination by particles according to ISO 4406 of hydraulic oil, the size threshold of ≥ 14 μm is more suitable for distinguishing various types of contaminant particles.
Raman spectroscopy (RS) combined with deep learning provides a noninvasive approach for breast cancer tissue diagnosis, but small clinical Raman datasets can limit classifier performance. This study examined how the number and class distribution of synthetic spectra generated by a generative adversarial network (GAN) affect breast tissue classification using a one-dimensional convolutional neural network (1D-CNN). The dataset included 49 real Raman spectra from formalin-fixed breast tissue samples, comprising 17 cancerous, 14 normal, and 18 benign spectra. A one-dimensional GAN generated synthetic spectra separately for each tissue class. Two augmentation strategies were assessed: a class-balanced strategy, in which each class was expanded to the same number of spectra, and a class-imbalanced strategy, in which the dataset was enlarged while preserving the original class proportions. Classifier performance was evaluated with leave-one-spectrum-out cross-validation, and results were averaged over at least 10 independent runs. In the conservative evaluation, synthetic spectra were added only to the training folds, while each of the 49 real spectra was held out for testing in turn. Under this setting, GAN-based augmentation substantially improved benign-class classification and increased overall accuracy compared with training on the original spectra alone. The findings indicate that GAN-based augmentation can improve RS-based breast cancer classification although performance gains saturate as the amount of synthetic data increases. The study also discusses limitations related to dataset size, formalin fixation, and the absence of sample-level grouping.
Glutamate (Glu) is a major excitatory neurotransmitter and an essential component of cerebral energy metabolism, both of which may be influenced by stress and anesthesia. This study investigated the temporal profile of hippocampal Glu in control mice and mice exposed to acute restraint stress during continuous anesthesia and after intervening periods of awake recovery. Twenty-four mice were assigned to control and stress groups (n = 12 per group), and hippocampal Glu was measured longitudinally using in vivo proton magnetic resonance spectroscopy at 0.33, 0.67, 1, 1.33, 1.67, 2, 4, 6, and 24 h. The first six measurements were obtained under uninterrupted isoflurane anesthesia, whereas measurements at 4, 6, and 24 h were obtained after intervening periods of awake recovery. Glu was designated a priori as the primary metabolite, and the continuous-anesthesia and recovery phases were analyzed separately using two-way mixed-design repeated-measures analyses of variance. During continuous anesthesia, Glu decreased significantly over time (p < 0.001), whereas neither the overall group effect (p = 0.733) nor the group-by-time interaction (p = 0.711) was significant. During the recovery phase, Glu was lower overall in the stress group than in the control group (p = 0.011), although the group-by-time interaction did not reach statistical significance (p = 0.059). Between-group differences observed at 4 and 24 h before multiplicity correction were no longer significant after Holm adjustment (adjusted p = 0.062 for both). In an exploratory analysis using the 2-h measurement as a post hoc reference, Glu increased from 2 to 4 h in the control group but decreased slightly in the stress group, resulting in a significant difference in change between the groups (Holm-adjusted p = 0.033). No group differences were detected for the 2 -> 6- or 2 -> 24-h changes. These findings indicate that hippocampal Glu varies substantially during prolonged anesthesia and suggest a possible stress-related alteration during the early postanesthetic period. Because the 2-h reference and subsequent change contrasts were not prespecified, however, the early difference should be considered exploratory and requires prospective confirmation in an independent, adequately powered study.
Ensuring the authenticity of soybean geographical origin is critical for food safety, quality control, and fair trade. In this study, a rapid, cost-effective, and highly interpretable strategy for soybean origin traceability was developed using laser-induced breakdown spectroscopy (LIBS) combined with chemometrics. LIBS spectra were collected from soybean samples covering five major producing regions in China, yielding a total of 650 qualified spectra for analysis. A dual-pathway feature engineering framework was implemented to systematically compare knowledge-driven selection (eight characteristic emission lines) and data-driven extraction (principal component analysis [PCA]). The PCA-linear discriminant analysis (LDA) model achieved 98.7% classification accuracy on an independent test set, significantly outperforming models based on expert-selected spectral lines. Six key elements (Ca, Mg, Na, K, Fe, and P) were identified as the primary discriminatory markers, contributing more than 85% of the feature importance, and their intensity variations were explicitly correlated with regional soil geochemistry, providing a transparent and physically interpretable classification mechanism. Compared with conventional analytical methods such as ICP-MS, the proposed LIBS-based protocol shortens analysis time from hours to minutes and can substantially reduce operational costs (with a preliminary estimated saving of approximately 70% based on routine laboratory cost accounting for consumables, instrument depreciation, and labor), while maintaining excellent precision with relative standard deviations below 5%. This work demonstrates that LIBS combined with interpretable chemometric models provides a reliable, practical, and physically interpretable analytical strategy that shows technical viability for rapid soybean origin authentication based on single-year samples. As a proof-of-concept study, multiyear sample datasets are required in future work to eliminate bias and overfitting risks, although the current results already indicate high potential for high-throughput screening and on-site monitoring in food control laboratories and across complex supply chains.
Honey is a natural product renowned for its nutritional and medicinal properties. However, the increasing market demand in Pakistan has led to frequent instances of adulterated and fraudulent honey. Conventional detection methods are widely used but are often ineffective in accurately identifying adulteration. This study proposes a novel approach to detecting adulteration in honey, specifically the mixing of sugar, using machine learning (ML) techniques in conjunction with hyperspectral imaging (HSI). By leveraging HSI, we capture distinct spectral features of both pure and adulterated honey. These spectral features are processed and analyzed using ML algorithms trained on a dataset comprising pure and mixed honey samples. Experiments involve binary, as well as, multiclass (5 classes) honey samples with different levels of adulteration. In addition, data balancing is also carried out using synthetic minority oversampling technique (SMOTE). The results prove the proposed approach to be efficient in detecting adulteration across various honey varieties, with linear regression achieved 99.9% while support vector machine and multilayer perceptron obtaining a 98.7% accuracy for binary class. For the multiple classes of honey, the multilayer perceptron shows a 96.67% accuracy outperforming the remaining ML and deep learning models. Among the deep learning models, the convolutional neural network (CNN) achieved the best performance with an accuracy of 94.67% after applying the SMOTE & times; 3 (three times of over samples of original samples of each class) data balancing strategy, whereas recurrent neural network (RNN), long short-term memory (LSTM), and gated recurrent unit (GRU) models showed comparatively lower performance.
