Unknown Samples Research Articles

Enhancing Open-World Bacterial Raman Spectra Identification by Feature Regularization for Improved Resilience against Unknown Classes.

The combination of deep learning techniques and Raman spectroscopy shows great potential offering precise and prompt identification of pathogenic bacteria in clinical settings. However, the traditional closed-set classification approaches assume that all test samples belong to one of the known pathogens, and their applicability is limited since the clinical environment is inherently unpredictable and dynamic, unknown, or emerging pathogens may not be included in the available catalogs. We demonstrate that the current state-of-the-art neural networks identifying pathogens through Raman spectra are vulnerable to unknown inputs, resulting in an uncontrollable false positive rate. To address this issue, first we developed an ensemble of ResNet architectures combined with the attention mechanism that achieves a 30-isolate accuracy of 87.8 ± 0.1%. Second, through the integration of feature regularization by the Objectosphere loss function, our model both achieves high accuracy in identifying known pathogens from the catalog and effectively separates unknown samples drastically reducing the false positive rate. Finally, the proposed feature regularization method during training significantly enhances the performance of out-of-distribution detectors during the inference phase improving the reliability of the detection of unknown classes. Our algorithm for Raman spectroscopy empowers the identification of previously unknown, uncataloged, and emerging pathogens ensuring adaptability to future pathogens that may surface. Moreover, it can be extended to enhance open-set medical image classification, bolstering its reliability in dynamic operational settings.

Identification research of chemical process leakage based on deep learning and correlation-distance graph coding

As a leak identification method in chemical processes, deep learning is limited by inadequate sample labels and highly coupled, nonlinear process variables, therefore difficult to model with a high identification accuracy. To address these problems, a fusion correlation-distance graph (CDG) coding aided deep learning method is proposed. First, the mechanism model is built based on dynamic process data with different leakage labels. Then, the weighted degree and spatial distribution of the dynamic dataset are computed through innovatively designed correlation and distance coding, aided by sliding window fusion into a correlation-distance matrix. Finally, the well-trained long short-term memory (LSTM) is constructed to learn the deep features of the leakage process and identify them. The applications in two industrial cases, CO2 capture and ammonia synthesis, show that CDG-LSTM presents state-of-the-art identification performance with R2 scores of 0.994 and 0.986. 6.6% and 18.4% improvement over LSTM. In addition, the proposed method effectively categorizes the unknown leak samples into known labels with nearby locations, providing a valuable reference for operators to determine the unknown leak locations.

Online vs Traditional Recruitment Approaches: Preferences of Saudi Employers and Employees

Purpose: This study analyses and compares traditional and e-recruitment methods for job applications from the perspectives of employees and employers. Research Design: The study collected data from employees and employers in business organizations located in Yanbu Industrial City, Kingdom of Saudi Arabia. An internet-mediated approach was used, and a survey questionnaire was designed. The sample size was determined to be 328 using an unknown population sample size formula. Data were gathered from 103 employees and 53 employers. Descriptive statistical analysis was conducted, and Chi-square was used to analyze the approaches of Traditional and e-recruitment for the job application process. Findings: The findings indicate that a high percentage of employees and employers’ preference for online methods in job applications. However, the chi-square test reveals a discrepancy between the perspectives of employees and employers. Employees support the use of online methods, while employers reject the hypothesis that online application methods are effective. Implications: One of the implications of online recruitment systems is that managers who use e-recruitment systems may have concerns about the accuracy and accountability of applicant data. Additionally, the lack of in-person interactions during the online application process can lead to frustration for job seekers and missed opportunities for employers to obtain or share more information. Originality: This study is arguably one of the leading studies that investigates the perceptions of employers and employees regarding the e-recruitment process in Saudi Arabia. Its findings would be of great help to employers in improving the e-recruitment system.

Machine Learning Model Stability for Sub-Regional Classification of Barossa Valley Shiraz Wine Using A-TEEM Spectroscopy.

With a view to maintaining the reputation of wine-producing regions among consumers, minimising economic losses caused by wine fraud, and achieving the purpose of data-driven terroir classification, the use of an absorbance-transmission and fluorescence excitation-emission matrix (A-TEEM) technique has shown great potential based on the molecular fingerprinting of a sample. The effects of changes in wine composition due to ageing and the stability of A-TEEM models over time had not been addressed, however, and the classification of wine blends required investigation. Thus, A-TEEM data were combined with an extreme gradient boosting discriminant analysis (XGBDA) algorithm to build classification models based on a range of Shiraz research wines (n = 217) from five Barossa Valley sub-regions over four vintages that had aged in bottle for several years. This spectral fingerprinting and machine learning approach revealed a 100% class prediction accuracy based on cross-validation (CV) model results for vintage year and 98.8% for unknown sample prediction accuracy when splitting the wine samples into training and test sets to obtain the classification models. The modelling and prediction of sub-regional production area showed a class CV prediction accuracy of 99.5% and an unknown sample prediction accuracy of 93.8% when modelling with the split dataset. Inputting a sub-set of the current A-TEEM data into the models generated previously for these Barossa sub-region wines yielded a 100% accurate prediction of vintage year for 2018-2020 wines, 92% accuracy for sub-region for 2018 wines, and 91% accuracy for sub-region using 2021 wine spectral data that were not included in the original modelling. Satisfactory results were also obtained from the modelling and prediction of blended samples for the vintages and sub-regions, which is of significance when considering the practice of wine blending.

Synergizing Deep Learning-Enabled Preprocessing and Human-AI Integration for Efficient Automatic Ground Truth Generation.

The progress of incorporating deep learning in the field of medical image interpretation has been greatly hindered due to the tremendous cost and time associated with generating ground truth for supervised machine learning, alongside concerns about the inconsistent quality of images acquired. Active learning offers a potential solution to these problems of expanding dataset ground truth by algorithmically choosing the most informative samples for ground truth labeling. Still, this effort incurs the costs of human labeling, which needs minimization. Furthermore, automatic labeling approaches employing active learning often exhibit overfitting tendencies while selecting samples closely aligned with the training set distribution and excluding out-of-distribution samples, which could potentially improve the model's effectiveness. We propose that the majority of out-of-distribution instances can be attributed to inconsistent cross images. Since the FDA approved the first whole-slide image system for medical diagnosis in 2017, whole-slide images have provided enriched critical information to advance the field of automated histopathology. Here, we exemplify the benefits of a novel deep learning strategy that utilizes high-resolution whole-slide microscopic images. We quantitatively assess and visually highlight the inconsistencies within the whole-slide image dataset employed in this study. Accordingly, we introduce a deep learning-based preprocessing algorithm designed to normalize unknown samples to the training set distribution, effectively mitigating the overfitting issue. Consequently, our approach significantly increases the amount of automatic region-of-interest ground truth labeling on high-resolution whole-slide images using active deep learning. We accept 92% of the automatic labels generated for our unlabeled data cohort, expanding the labeled dataset by 845%. Additionally, we demonstrate expert time savings of 96% relative to manual expert ground-truth labeling.

Raman Match: Application for Automated Identification of Minerals from Raman Spectroscopy Data

Abstract Raman spectroscopy is a rapid, nondestructive analysis technique that is used in various scientific disciplines, including chemistry, materials science, and biology. The analysis of Raman spectra and the identification of specific substances in unknown samples can be complex and time-consuming tasks due to the large database of Raman spectra. The Raman Match application was developed to simplify and automate the sample identification process through a search and match method. The application integrates the well-established RRUFF Raman database with the Python programming language and provides a user-friendly graphical interface to load Raman spectra, identify and fit peaks, match peaks to the reference libraries, visualize the results, and generate publication-ready figures. The application offers a swift and automated method for mineral identification using Raman spectroscopy in both laboratory and field settings, as well as during planetary exploration missions to extraterrestrial environments with constraints on time and resources.

A Label-Free Droplet Sorting Platform Integrating Dielectrophoretic Separation for Estimating Bacterial Antimicrobial Resistance.

Antimicrobial resistance (AMR) has become a crucial global health issue. Antibiotic-resistant bacteria can survive after antibiotic treatments, lowering drug efficacy and increasing lethal risks. A microfluidic water-in-oil emulsion droplet system can entrap microorganisms and antibiotics within the tiny bioreactor, separate from the surroundings, enabling independent assays that can be performed in a high-throughput manner. This study presents the development of a label-free dielectrophoresis (DEP)-based microfluidic platform to sort droplets that co-encapsulate Escherichia coli (E. coli) and ampicillin (Amp) and droplets that co-encapsulate Amp-resistant (AmpR) E. coli with Amp only based on the conductivity-dependent DEP force (FDEP) without the assistance of optical analyses. The 9.4% low conductivity (LC) Luria-Bertani (LB) broth diluted with 170 mM mannitol can maintain E. coli and AmpR E. coli growth for 3 h and allow Amp to kill almost all E. coli, which can significantly increase the LCLB conductivity by about 100 μS/cm. Therefore, the AmpR E. coli/9.4%LCLB/Amp where no cells are killed and the E. coli/9.4%LCLB/Amp-containing droplets where most of the cells are killed can be sorted based on this conductivity difference at an applied electric field of 2 MHz and 100 Vpp that generates positive FDEP. Moreover, the sorting ratio significantly decreased to about 50% when the population of AmpR E. coli was equal to or higher than 50% in droplets. The conductivity-dependent DEP-based sorting platform exhibits promising potential to probe the ratio of AmpR E. coli in an unknown bacterial sample by using the sorting ratio as an index.

VOC data-driven evaluation of vehicle cabin odor: from ANN to CNN-BiLSTM.

Emissions of volatile organic compounds (VOCs) in vehicles represent a significant problem, causing unpleasant odors. To mitigate VOCs and odors in vehicles, it is critical to choose interior parts with low odor and VOC emissions. However, prevailing odor evaluation methods are subjective, costly, and potentially harmful to the health of evaluators. In this study, we analyzed 139 automotive interior parts and 92 vehicles, establishing a cost-effective, data-driven method for odor evaluation. The contents of benzene, toluene, ethylbenzene, xylene, styrene, formaldehyde, acetaldehyde, acrolein, and total volatile organic compounds (TVOC) were detected by thermal desorption gas chromatography-mass spectrometry (TD-GC/MS) and high-performance liquid chromatography with an ultraviolet detector (HPLC-UV). Professional odor evaluators assessed the odors, identifying intensity levels from 2.0 to 4.5 in interior parts and 2.5 to 3.5 in whole vehicles. Leveraging this data, we applied four supervised learning algorithms to develop predictive models for the odor intensity of both interior parts and entire vehicles. During model training, we implemented early stopping techniques for the artificial neural network (ANN) and convolutional neural network-bidirectional long short-term memory (CNN-BiLSTM) models, while optimizing the support vector machine (SVM) and extreme gradient boosting (XGBoost) models using the GridSearch algorithm. The evaluation results reveal that the CNN-BiLSTM model performs the best, achieving an average accuracy of 89% for unknown samples within an odor intensity level of 0.5. The root mean square error (RMSE) is 0.24, and the mean absolute error (MAE) is 0.08. The model also underwent a sevenfold cross-validation, achieving an accuracy of 83.43%. Additionally, we employed SHapley Additive exPlanations (SHAP) for the interpretative analysis of the model, which confirmed the consistency of each VOC's odor contribution with human olfactory rules. By predicting odors based on VOCs through supervised learning, this study reduces the costs and enhances the efficiency and applicability of odor assessment across various vehicle interiors.

Species discrimination from hair using ATR-FTIR spectroscopy: Application in wildlife forensics

Hair is a commonly encountered trace evidence in wildlife crimes involving mammals and can be used for species identification which is essential for subsequent judicial proceedings. This proof of concept study aims, to distinguish the black guard hair of three wild cat species belonging to the genus Panthera i.e. Royal Bengal Tiger (Panthera tigris tigris), Indian Leopard (Panthera pardus fusca), and Snow Leopard (Panthera uncia) using a rapid and non-destructive ATR-FTIR spectroscopic technique in combination with chemometrics. A training dataset including 72 black guard hair samples of three species (24 samples from each species) was used to construct chemometric models. A PLS2-DA model successfully classified these three species into distinct classes with R-Square values of 0.9985 (calibration) and 0.8989 (validation). VIP score was also computed, and a new PLS2DA-V model was constructed using variables with a VIP score ≥ 1. External validation was performed using a validation dataset including 18 black guard hair samples (6 samples per species) to validate the constructed PLS2-DA model. It was observed that PLS2-DA model provides greater accuracy and precision compared to the PLS2DA-V model during cross-validation and external validation. The developed PLS2-DA model was also successful in differentiating human and non-human hair with R-Square values of 0.99 and 0.91 for calibration and validation, respectively. Apart from this, a blind test was also carried out using 10 unknown hair samples which were correctly classified into their respective classes providing 100 % accuracy. This study highlights the advantages of ATR-FTIR spectroscopy associated with PLS-DA for differentiation and identification of the Royal Bengal Tiger, Indian Leopard, and Snow Leopard hairs in a rapid, accurate, eco-friendly, and non-destructive way.

Quantification of total human DNA using qPCR in mixture samples of Crimes Against Sexual Freedom to estimate the success of analysis with autosomal STRs and Y-STRs

ABSTRACT In this study, 241 mixed autosomal STR genetic profiles (swabs − 172, slides − 69) from processed casework samples were analysed. From these samples, DNA was extracted and quantified by qPCR; the results were analysed with the PowerQuant Analysis Software. The extracted DNA samples were amplified with the Investigator 24Plex QS and VeriFiler kits and separated and detected in an ABI 3500 Genetic Analyzer. A quantification correlation was observed between the total autosomal DNA targets [Auto] and the Y chromosome target [Y] in mixed samples in which the success of amplification of the male autosomal STRs markers was associated with high values of the quantification of the target of [Y] and low DNA ratio values [Auto]/[Y] (p = 0.00 < 0.05). Samples with [Auto]/[Y] ratio values much higher than the estimated value required analysis with Y-STR markers. The study herein, which assessed a relatively large number of sexual assault samples, demonstrates that quantitation ratio values of DNA mixtures effectively contribute to the decision-making process on whether and what genetic marker system should be selected for analysis of a male component in unknown casework samples.

Machine learning assists the sensor array constructed by the tri-emission carbon dots to detect multiple metal ions

Metal ions are related to the health of daily life, which urges to be detected and identified by efficient sensors. We first synthesize tri-emission carbon dots for the functional determination of six metal ions and obtain the multi-dimension data to construct the fluorescent sensor array in the work. Inspired by the data-driven method, machine learning is employed in a sensor array to detect and analyze the six metal ions. The ion classification and concentration models are constructed to identify metal ion types and corresponding concentrations. The classification model could correctly discriminate six metal ions, reaching 99.5 % accuracy and the concentration model exhibits an excellent performance range from 30 to 330 μM, with a mean concentration difference value of 0.12 μM and R2 value of 0.9845 for the unknown sample. Furthermore, six metal ions in tap water also can be distinguished with 96 % accuracy and the recovery rate is between 101.89 % and 106.37 %. This work shows that the machine learning combined sensor array has significant advantages in detecting multiple metal ions and provides a novel thought for analyzing sensing data, accelerating the development of data-driven strategies in the sensing field. © 2001 Elsevier Science. All rights reserved.

Comparison of laser-induced breakdown spectroscopy (LIBS) and ICP analysis results for measuring Pb and Zn in soil

Context Laser-induced breakdown spectroscopy (LIBS) is a rapid, multielement analytical technique. It is particularly suitable for the qualitative and quantitative analyses of heavy metals in solid samples. Aims To validate the technique, the LIBS data were compared with the data obtained via conventional inductively coupled plasma (ICP) spectroscopy for the same soil samples. Methods In this study, standard and unknown soil samples from contaminated areas were prepared and fixed to an adhesive tape for LIBS analysis. The soils were also digested with acids for ICP analysis. The emission intensity of one selected line for each of the two analytes, i.e. lead (Pb) and zinc (Zn), was normalised to the background signal and plotted as a function of the concentration values previously determined via ICP analysis. Key results The data demonstrated good linearity for the calibration lines drawn, and the correlation between the ICP and LIBS data was confirmed by the satisfactory agreement between the corresponding values. Conclusions The concentration coefficient of determination (R2) between LIBS and ICP-aqua regia digestion analysis or ICP-total digestion analysis were &gt;0.86 and &gt;0.89 for Pb and Zn, respectively. The total analysis time for the LIBS method was 310 min, which was 54.40% shorter than that for the ICP method (680 min). Implications Consequently, LIBS can be used to measure Pb and Zn in soils without any chemical preparation.

Seabream Quality Monitoring Throughout the Supply Chain Using a Portable Multispectral Imaging Device

Monitoring food quality throughout the supply chain in a rapid and cost-effective way allows on-time decision−making, reducing food waste, and increasing sustainability. A portable multispectral imaging sensor was used for the rapid prediction of microbiological quality of fish fillets. Seabream fillets, packaged either in aerobic or vacuum conditions, were collected from both aquaculture and retail stores, while images were also acquired both from the skin and the flesh side of the fish fillets. In parallel to image acquisition, the microbial quality was also estimated for each fish fillet. The data were used for the training of predictive artificial neural network (ANN) models for the estimation of total aerobic counts (TACs). Models were built separately for fish parts (i.e., skin, flesh) and packaging conditions and were validated using two approaches (i.e., validation with data partitioning and external validation using samples from retail stores). The performance of the ANN models for the validation set with data partitioning was similar for the data collected from the flesh (RMSE = 0.402–0.547) and the skin side (RMSE = 0.500–0.533) of the fish fillets. Similar performance also was obtained from validation of the models of the different packaging conditions (i.e., aerobic, vacuum). The prediction capability of the models combining both air and vacuum packaged samples (RMSE = 0.531) was slightly lower compared to the models trained and validated per packaging condition, individually (RMSE = 0.510, 0.516 in air and vacuum, respectively). The models tested with unknown samples (i.e., fish fillets from retail stores-external validation) showed poorer performance (RMSE = 1.061–1.414) compared to the models validated with data partitioning (RMSE = 0.402–0.547). Multispectral imaging sensor appeared to be efficient for the rapid assessment of the microbiological quality of fish fillets for all the different cases evaluated. Hence, these outcomes could be beneficial not only for the industry and food operators but also for the authorities and consumers.

An intelligent indicating label using mulberry pigment, polylactic acid and rutile TiO2 nanoparticle for monitoring the freshness of cold fresh pork

The purpose of this work is to develop a new type of intelligent indicating label with high sensitivity to volatile bioamine, anti-ultraviolet light and anti-pigment leakage characteristics by electrospinning for real-time visual monitoring the freshness of cold fresh pork. By investigating pH discoloration sensitivity and light stability of plant pigments with different structures, mulberry pigment was selected as the indicator of indicating label. The content of label components (mulberry pigment, polylactic acid and TiO2 nanoparticle) significantly affected label performance (light stability, responsiveness to trimethylamine, and anti-leakage of mulberry pigment), and the optimal component content was determined to be 6.01% (w/w) of polylactic acid, 0.62% (w/w) of TiO2 nanoparticles and 0.29% (w/w) of mulberry pigment by response surface test based on label performance as response value. A close correlation (the correlation coefficient R values of 0.86–0.97) between the total color difference (ΔE) of the label and the freshness indexes (total plate count, total volatile basic nitrogen content, pH, thiobarbituric acid value and sensory score) of cold fresh pork during cold storage was revealed, and the freshness level of unknown pork samples could be well quantitatively predicted by partial least squares regression models with the determination coefficient (R2) values of 0.89–0.92, facilitating producers, dealers and consumers to know pork quality promptly.

Adulterants present in the San Diego county fentanyl supply: a laboratory analysis of seized law enforcement samples

BackgroundThe opioid overdose crisis is one of the worst public health crises ever to face the US and emerging evidence suggests its effects are compounded by the presence of drug adulterants. Here we report our efforts to characterize the adulterants present within the local fentanyl supply of San Diego County, obtained from undifferentiated drug samples seized by local law enforcement over the calendar year 2021.MethodsThirty-two participating local law enforcement agencies across San Diego submitted 4838 unknown individual illicit drug samples (total of 312 kg) to the San Diego County Sheriff’s Department Regional Crime Laboratory for identification.ResultsQualitative analysis of these samples via FTIR and GC-MS identified methamphetamine (38.7%), fentanyl (20.8%), diacetylmorphine (heroin) (10.2%), codeine (5.8%) and alprazolam (4.3%) as the most common illicit substances and the presence of 52 unique adulterants. The most common adulterants included 4-methylaminoantipyrine (4-MAAP) (10.9%), mannitol (9%), acetaminophen (8.5%), methamphetamine (4.2%), diacetylmorphine (heroin) (3.6%), tramadol (1.9%), and xylazine (1.7%). Several additional pharmacologically active adulterants and contaminants of interest were also identified.ConclusionThis analysis is vital for public health use and harm reduction efforts at the level of the individual consumer. Continued direct surveillance of the drug supply is necessary for the detection of potentially harmful adulterants that may pose serious threats to the public.

Nanopore sequencing in distinguishing between wild-type and vaccine strains of Varicella-Zoster virus

BackgroundThe introduction of varicella vaccines into routine pediatric immunization programs has led to a considerable reduction in varicella incidence. However, there have been reports of varicella, herpes zoster, and meningitis caused by the vaccine strain of varicella-zoster virus (VZV), raising concerns. Establishing the relationship between the wild-type and vaccine strains in VZV infections among previously vaccinated individuals is crucial. Differences in the single nucleotide polymorphisms (SNPs) among vaccine strains can be utilized to identify the strain. In this study, we employed nanopore sequencing to identify VZV strains and analyzed clinical samples. MethodsWe retrospectively examined vesicle and cerebrospinal fluid samples from patients with VZV infections. One sample each of the wild-type and vaccine strains, previously identified using allelic discrimination real-time PCR and direct sequencing, served as controls. Ten samples with undetermined VZV strains were included. After DNA extraction, a long PCR targeting the VZV ORF62 region was executed. Nanopore sequencing identified SNPs, allowing discrimination between the vaccine and wild-type strains. ResultsNanopore sequencing confirmed SNPs at previously reported sites (105,705, 106,262, 107,136, and 107,252), aiding in distinguishing between wild-type and vaccine strains. Among the ten unknown samples, nine were characterized as wild strains and one as a vaccine strain. Even in samples with low VZV DNA levels, nanopore sequencing was effective in strain identification. ConclusionThis study validates that nanopore sequencing is a reliable method for differentiating between the wild-type and vaccine strains of VZV. Its ability to produce long-read sequences is remarkable, allowing simultaneous confirmation of known SNPs and the detection of new mutations. Nanopore sequencing can serve as a valuable tool for the swift and precise identification of wild-type and vaccine strains and has potential applications in future VZV surveillance.

Open Set Bearing Fault Diagnosis with Domain Adaptive Adversarial Network under Varying Conditions

Bearing fault diagnosis is a pivotal aspect of monitoring rotating machinery. Recently, numerous deep learning models have been developed for intelligent bearing fault diagnosis. However, these models have typically been established based on two key assumptions: (1) that identical fault categories exist in both the training and testing datasets, and (2) the datasets used for testing and training are assumed to follow the same distribution. Nevertheless, these assumptions prove impractical and fail to accurately depict real-world scenarios, particularly those involving open-world assumption fault diagnosis in multi-condition scenarios. For that purpose, an open set domain adaptive adversarial network framework is proposed. Specifically, in order to improve the learning of distribution characteristics in different fields, comprehensive training is implemented using a deep convolutional autoencoder model. Additionally, to mitigate the negative transfer resulting from unknown fault samples in the target domain, the similarity of each target domain sample and the shared classes in the source domain are estimated using known class classifiers and extended classifiers. Similarity weight values are assigned to each target domain sample, and an unknown boundary is established in a weighted manner. This approach is employed to establish the alignment between the classes shared between the two domains, enabling the classification of known fault classes, while allowing the recognition of unknown fault classes in the target domain. The efficacy of our suggested approach is empirically validated using different datasets.

Research on automotive scrap metal classification method using laser-induced breakdown spectroscopy and two-step clustering algorithm

In the recycling of scrap metal, the establishment of the classification database of recyclables has the advantages of fast classification speed and high analysis accuracy. However, the classification and recycling of unknown samples become highly significant due to the extensive variety of standard metal samples and the challenges in obtaining them. In this study, a method for multi-element classification of automotive scrap metals in general environmental conditions was achieved by utilizing laser-induced breakdown spectroscopy (LIBS) and two-step clustering algorithm (K-means, hierarchical clustering). The two unsupervised learning algorithms were employed to cluster the LIBS spectral data of 60 automotive scrap metal samples rapidly and hierarchically. Three rare metal elements and three elements for distinguishing metal categories were selected to meet the recycling requirements. After applying the multiplicative scatter correction to the spectral data for calibration, the initial clustering clusters were determined using the Davies–Bouldin index, Calinski–Harabasz index, and silhouette coefficient. Then, the Kruskal–Wallis test was conducted on each cluster to check the significance. The clusters that failed the test were split and reclustered until all clusters met the significance criterion (α=0.05). The accuracy of the proposed method for classifying the collected automotive scrap metals reached 97.6%. This indicates the great potential of this method in the field of automotive scrap metal classification.

An auto-regulated universal domain adaptation network for uncertain diagnostic scenarios of rotating machinery

In recent years, domain adaptation techniques have garnered significant attention in the field of intelligent fault diagnosis for mechanical equipment. Domain adaptation techniques can effectively enable the reuse of diagnostic knowledge, thereby improving the generalization performance of the trained model. However, many intelligent diagnostic models are tailored to specific diagnostic scenarios, including closed-set domain adaptation, partial domain adaptation, and open-set domain adaptation. When these models built for specific scenarios are applied to other scenarios, their diagnostic performance will deteriorate catastrophically. This study proposes an auto-regulated universal domain adaptation network (AUDAN) for uncertain diagnostic scenarios of rotating machinery, which can handle multiple diagnostic scenarios and adaptively set threshold boundaries for each fault category based on data distribution structure to distinguish known and unknown samples. Specifically, a novel integration loss function is designed to train the proposed AUDAN. The self-adaptive reweight module is introduced into the proposed method to distribute the relative weight for each loss function automatically. Experiments on two rotating machinery datasets are carried out for validations. The experimental results demonstrate that the proposed AUDAN performs effectively in uncertain diagnostic scenarios and is promising in addressing the fault classification tasks in real industries.

Entropic Open-Set Active Learning

Active Learning (AL) aims to enhance the performance of deep models by selecting the most informative samples for annotation from a pool of unlabeled data. Despite impressive performance in closed-set settings, most AL methods fail in real-world scenarios where the unlabeled data contains unknown categories. Recently, a few studies have attempted to tackle the AL problem for the open-set setting. However, these methods focus more on selecting known samples and do not efficiently utilize unknown samples obtained during AL rounds. In this work, we propose an Entropic Open-set AL (EOAL) framework which leverages both known and unknown distributions effectively to select informative samples during AL rounds. Specifically, our approach employs two different entropy scores. One measures the uncertainty of a sample with respect to the known-class distributions. The other measures the uncertainty of the sample with respect to the unknown-class distributions. By utilizing these two entropy scores we effectively separate the known and unknown samples from the unlabeled data resulting in better sampling. Through extensive experiments, we show that the proposed method outperforms existing state-of-the-art methods on CIFAR-10, CIFAR-100, and TinyImageNet datasets. Code is available at https://github.com/bardisafa/EOAL.