Labeling Errors Research Articles

Intelligent Fault Diagnosis With Noisy Labels via Semisupervised Learning on Industrial Time Series

Deep neural networks (DNNs) excel at industrial fault diagnosis. Their performance heavily relies on the quality of human-annotated labels. Due to perception limitations of annotators, industrial time series samples (such as vibration and voltage signals) are frequently mislabeled in several conditions, such as samples with frequency domain feature differences and samples on class borders. Hence, an annotated industrial dataset will inevitably contain noisy labels at a certain level, leading to over-fitting and poor generalization of DNNs. In this work, we introduce an Industrial Noisy Label Semi-Supervised Learning (INL-SSL) fault diagnosis approach, addressing the problem that a certain number of samples in an industrial dataset are mislabeled. The proposed INL-SSL architecture simultaneously trains two DNNs, which cross-train on each other to filter noisy label errors. In particular, a fitted Gaussian mixture model divides time series samples of each DNN flow into an unlabeled set with samples likely to be noisy and a labeled set with samples likely to be clean. Given the labeled and unlabeled data, we proposed a time series MixMatch semi-supervised learning strategy to train the diagnostic model. Ablation study verifies the benefit of the proposed time series augmentation techniques for semi-supervised training. Extensive experiments on a benchmark industrial dataset of rolling element bearings (REB) reveal that the INL-SSL outperforms state-of-the-art approaches. On another self-collected REB dataset, the proposed approach also exceeds other comparison methods under noise ratios from 20% to 90%, validating the model's generalizability.

Annotation-efficient training of medical image segmentation network based on scribble guidance in difficult areas.

The training of deep medical image segmentation networks usually requires a large amount of human-annotated data. To alleviate the burden of human labor, many semi- or non-supervised methods have been developed. However, due to the complexity of clinical scenario, insufficient training labels still causes inaccurate segmentation in some difficult local areas such as heterogeneous tumors and fuzzy boundaries. We propose an annotation-efficient training approach, which only requires scribble guidance in the difficult areas. A segmentation network is initially trained with a small amount of fully annotated data and then used to produce pseudo labels for more training data. Human supervisors draw scribbles in the areas of incorrect pseudo labels (i.e., difficult areas), and the scribbles are converted into pseudo label maps using a probability-modulated geodesic transform. To reduce the influence of the potential errors in the pseudo labels, a confidence map of the pseudo labels is generated by jointly considering the pixel-to-scribble geodesic distance and the network output probability. The pseudo labels and confidence maps are iteratively optimized with the update of the network, and the network training is promoted by the pseudo labels and the confidence maps in turn. Cross-validation based on two data sets (brain tumor MRI and liver tumor CT) showed that our method significantly reduces the annotation time while maintains the segmentation accuracy of difficult areas (e.g., tumors). Using 90 scribble-annotated training images (annotated time: ~ 9h), our method achieved the same performance as using 45 fully annotated images (annotation time: > 100h) but required much shorter annotation time. Compared to the conventional full annotation approaches, the proposed method significantly saves the annotation efforts by focusing the human supervisions on the most difficult regions. It provides an annotation-efficient way for training medical image segmentation networks in complex clinical scenario.

A deep learning solution for particle size analysis in low resolution inline microscopy images based on generative adversarial network

We introduce a new strategy for image analysis of inline microscopy monitoring estimate particle size distribution using deep learning. The proposed method consists of two major components: First, a novel way to generate training image-label pairs with a high-level of credibility via a Cycle-consistent Generative Adversarial Network (CycleGAN), and second, a Mask-RCNN model trained with the generated data for the particle detection task. The proposed methodology eliminates the need for manual labeling in the training phase which is a labor-intensive step and can result in labeling errors given the fuzziness of these images. We studied the application of this strategy to images acquired with a particle vision and measurement (PVM) probe.The proposed methodology was applied to images of two particle morphologies with different sizes and concentrations. Our results showed that the proposed methodology could be inexpensively used to determine qualitative trends between crystal size distributions. This trend information is a very important aspect of crystallization process monitoring and is often enough to determine what is controlling the crystallization. Therefore, we see our approach as a step in the right direction to provide insights into the particularly challenging PVM inline microscopy monitoring process without the need for offline sampling.

Rethinking PASCAL-VOC and MS-COCO dataset for small object detection

The data and the algorithm are critical to deep learning-based small object detectors. In this paper, we rethink the PASCAL-VOC and MS-COCO dataset for small object detection. By visual analysis of the original annotations, we find that there are different labeling errors in these two datasets. To solve these problems, we build specific datasets, including SDOD, Mini6K, Mini2022 and Mini6KClean. The experimental results of several typical algorithms (e.g. SSD, YOLOv5, Faster RCNN and Deformable DETR) on the datasets show that data labeling errors (such as missing labels, category label errors, inappropriate labels) are another factor that affects the detection performance of small objects.

Machine Learning and Explainable Artificial Intelligence Using Counterfactual Explanations for Evaluating Posture Parameters.

Postural deficits such as hyperlordosis (hollow back) or hyperkyphosis (hunchback) are relevant health issues. Diagnoses depend on the experience of the examiner and are, therefore, often subjective and prone to errors. Machine learning (ML) methods in combination with explainable artificial intelligence (XAI) tools have proven useful for providing an objective, data-based orientation. However, only a few works have considered posture parameters, leaving the potential for more human-friendly XAI interpretations still untouched. Therefore, the present work proposes an objective, data-driven ML system for medical decision support that enables especially human-friendly interpretations using counterfactual explanations (CFs). The posture data for 1151 subjects were recorded by means of stereophotogrammetry. An expert-based classification of the subjects regarding the presence of hyperlordosis or hyperkyphosis was initially performed. Using a Gaussian progress classifier, the models were trained and interpreted using CFs. The label errors were flagged and re-evaluated using confident learning. Very good classification performances for both hyperlordosis and hyperkyphosis were found, whereby the re-evaluation and correction of the test labels led to a significant improvement (MPRAUC = 0.97). A statistical evaluation showed that the CFs seemed to be plausible, in general. In the context of personalized medicine, the present study's approach could be of importance for reducing diagnostic errors and thereby improving the individual adaptation of therapeutic measures. Likewise, it could be a basis for the development of apps for preventive posture assessment.

Evaluation of errors related to surgical pathology specimens of different hospital departments with a patient safety approach: a case study in Iran

BackgroundMost surgical specimen errors occur in the pre-analysis stage, which can be prevented. This study aims to identify errors related to surgical pathology specimens in one of the most comprehensive healthcare centers in Northeast Iran.MethodsThe present study is descriptive and analytical research conducted cross-sectionally in 2021 at Ghaem healthcare center in the Mashhad University of Medical Sciences on the basis of a census sampling. We used a standard checklist to collect information. Professors and pathologists evaluated the validity and reliability of the checklist using Cronbach’s alpha calculation method of 0.89. We analyzed the results using statistical indices, SPSS 21 software, and the chi-square test.ResultsOut of 5617 pathology specimens studied, we detected 646 errors. The highest number of errors is the mismatch of the specimen with the label (219 cases; 3.9%) and the non-compliance of the patient’s profile in the specimen sent with the label (129 cases; 2.3%), and the lowest errors are the inappropriate volume of the fixator(24 cases; 0.4%), and they accounted for insufficient sample size (25 cases; 0.4%). Based on Fisher’s exact test results, there was a significant difference between the proportion of errors in different departments and months.ConclusionConsidering the frequency of labeling errors in the stage before the analysis in the pathology department, the use of barcode imprinted in specimen containers, the removal of the paper request for pathology, the use of radio frequency chip technology, the use of the rechecking system and improving communication in different departments can be effective in reducing these errors.

487 Understanding the Medicines Labelling System in a Tertiary Hospital in Kuwait Using Work Domain Analysis

Abstract Introduction A report by the World Health Organisation highlighted a lack of clarity in medicine labelling regulations in Kuwait and the subsequent risk of patient harm from labelling errors (1). In response, the Undersecretary of the Kuwait Ministry of Health (MOH) issued a ministerial decree governing the labelling of medicines (M.D 4451/2021, June 2021) (2), stipulating a requirement to affix an identification label containing specific patient, medicine, and supplier information on medication packaging. Aim To develop a map of the Medicines Labelling System (MLS) in a tertiary hospital in Kuwait (TH). Methods A Work Domain Analysis (WDA) was conducted using qualitative data from 1) semi-structured interviews with a purposive sample of 10 nurses, 10 patients/caregivers, and 10 pharmacists involved in the delivery or administration of medicines to TH patients, as well as 2) an analysis of medicine labelling documents published by the Kuwaiti MOH and TH. After gaining informed consent, one-to-one semi-structured interviews were conducted in TH, in accordance with the latest COVID-19 prevention procedures. All interviews were audio-recorded and transcribed verbatim. Both deductive – informed by Abstraction Hierarchy (AH), the modelling tool of WDA – and inductive thematic analysis was conducted. Results Thirty participants took part in semi-structured interviews averaging 34 minutes, from which five levels of abstraction in the AH were constructed. The first level, “functional purposes”, describes the reason why any given system exists: for the MLS in TH, they were “compliance with regulations, policies, and protocols of the Kuwaiti MOH and TH” and “providing an identification of medicine issued by ISH pharmacies”. Five aspects of the second level, “values and priority measures”, were identified, including “optimisation of service workflow” and “time-efficient clinical practices.” The third level, “purpose-related functions”, describes the key functions that need to be in place for the MLS to work; these were “the production of medicines labels”, “coordination”, and “management of medicines and resources”. The last two levels in the AH, “object-related functions” and “physical objects”, describe the resources that are required in the MLS, their functionality, and their limitations: there were 16 groups of physical objects, which afforded 19 different functions. Means-ends links showed the crucial relationships and significance between functions at different levels of the AH, the most prominent of which were coordination and communication. Conclusion TH’s MLS is a complex system that was visually represented using an AH, showing the constraints and structure of the work system. The AH constructed a knowledge base. Subsequent analyses of the current system issues may draw on this knowledge, such as a potential analysis of how specific tasks, like the use of labels by nurses within wards, are performed. The limitations of this study included only exploring ‘work as imagined’ (i.e., accounts from healthcare staff and documents), which does not fully capture how the MLS functions on a day-to-day basis. Future work will focus on deepening our understanding of the MLS through observations of work in practice.

Toward structuring real-world data: Deep learning for extracting oncology information from clinical text with patient-level supervision

Most detailed patient information in real-world data (RWD) is only consistently available in free-text clinical documents. Manual curation is expensive and time consuming. Developing natural language processing (NLP) methods for structuring RWD is thus essential for scaling real-world evidence generation. We propose leveraging patient-level supervision from medical registries, which are often readily available and capture key patient information, for general RWD applications. We conduct an extensive study on 135,107 patients from the cancer registry of a large integrated delivery network (IDN) comprising healthcare systems in five western US states. Our deep-learning methods attain test area under the receiver operating characteristic curve (AUROC) values of 94%-99% for key tumor attributes and comparable performance on held-out data from separate health systems and states. Ablation results demonstrate the superiority of these advanced deep-learning methods. Error analysis shows that our NLP system sometimes even corrects errors in registrar labels.

AI-Hub 국내 신호등 데이터셋의 표본추출 기반 주석 오류 분석

본 논문은 AI-Hub 신호등 데이터셋에서 추출한 표본의 정밀한 분석을 통해 주석오류를 분류하는 방법을 제시하고, 주석생성 작업자의 주석 오류와 검증 시스템 오류를 구분함으로써 데이터셋 정제 방향을 제시한다. 또한, 데이터셋의 주석오류가 SSD와 YOLOv4의 객체 검출 정확도에 미치는 영향을 실험적으로 검증한다. This paper presents a labeling-error classification through the precise sampling-based analysis of the AI-Hub traffic light datasets, and suggests the direction of dataset refinement by distinguishing the crowdsourcing worker’s labeling errors from those by the verification system. We also experimentally validate the impacts of labeling errors in the dataset on the mAP of two representative object detection networks: SSD-512 and Scaled-YOLOv4-P6.

Improving robustness of industrial object detection by automatic generation of synthetic images from CAD models

AbstractObject detection (OD) is used for visual quality control in factories. Images that compose training datasets are often collected directly from the production line and labeled with bounding boxes manually. Such data represent well the inference context but might lack diversity, implying a risk of overfitting. To address this issue, we propose a dataset construction method based on an automated pipeline, which receives a CAD model of an object and returns a set of realistic synthetic labeled images (code publicly available). Our approach can be easily used by non‐expert users and is relevant for industrial applications, where CAD models are widely available. We performed experiments to compare the use of datasets obtained by the two different ways—collecting and labeling real images or applying the proposed automated pipeline—in the classification of five different industrial parts. To ensure that both approaches can be used without deep learning expertise, all training parameters were kept fixed during these experiments. In our results, both methods were successful for some objects but failed for others. However, we have shown that the combined use of real and synthetic images led to better results. This finding has the potential to make industrial OD models more robust to poor data collection and labeling errors, without increasing the difficulty of the training process.

Photovoltaics literature survey (no. 182)

Photovoltaics literature survey (no. 182)

AI’s 10 to Watch, 2022

AI’s 10 to Watch, 2022

Implicit Multimodal Crowdsourcing for Joint RF and Geomagnetic Fingerprinting

In fingerprint-based indoor localization, fusing radio frequency (RF) and geomagnetic signals has been shown to achieve promising results. To efficiently collect fingerprints, implicit crowdsourcing can be used, where signals sampled by pedestrians are automatically labeled with their locations on a map. Previous work on crowdsourced fingerprinting is often based on a single signal, which is susceptible to signal bias and labeling error. We study, for the first time, implicit multimodal crowdsourcing for joint RF and geomagnetic fingerprinting. The scheme, termed UbiFin, exploits the spatial correlation among RF, geomagnetic, and motion signals to mitigate the impact of sensor noise, leading to highly accurate and robust fingerprinting without the need for any explicit manual intervention. Using clustering and dynamic programming, UbiFin correlates spatially different signals and filters effectively mislabeled signals. We conduct extensive experiments on our campus and a large multi-story shopping mall. Efficient and simple to implement, UbiFin outperforms other state-of-the-art crowdsourcing schemes to construct RF and geomagnetic fingerprints in terms of accuracy and robustness (cutting fingerprint error by 40 percent in general).

Authentication and discrimination of tissue origin of bovine gelatin using combined supervised pattern recognition strategies

Gelatin is a protein-based product derived from bovine and porcine bone and skin. Gelatin is increasingly used in food and non-food products, but there are errors in labeling concerning the origin of commercial gelatin raw materials. These mislabeling may negatively impact religious or safety communities. In gelatin studies, species identification is common, but determining the origin of the tissues, bone and skin, is also crucial. The current study demonstrated the use of Raman spectroscopy as a rapid and economical method coupled with chemometrics to differentiate tissue origins of bovine gelatin. The proposed technique could also detect adulteration of bone origin gelatin with skin gelatin with 100% sensitivity and specificity. Accordingly, Raman spectroscopy, along with DD- SIMCA and PLS-DA methods, provides a reliable technique that can easily be applied to detecting the tissue source of bovine gelatin. Also, two- step classification method with goal of finding a suitable authentication of target classes have been applied. dd-SIMCA as class modeling for authentication and PLS-DA for discrimination of groups have been combined in this strategy.

On the revision of historical collections of three-dimensional models of foraminifers from the Paleontological Museum of St Petersburg State University

Two collections of plaster models of foraminifera from the Paleontological Museum of St. Petersburg State University were studied. The first one (No. ПЛ ОФ 721) is represented by 95 objects. The time of its acquisition and authorship were unknown, some of the models were kept without labels, there was no catalog and records of its receipt. The collection has not been used for teaching or exposure in the museum. As it turned out the creator of the original collection was Academician Alcide d'Orbigny, who first described foraminifers as a separate group of animals in 1826 in the monograph “Tableau méthodique de la classe des Céphalopodes”. The 100 plaster foraminiferal models were made by him to illustrate the classification the new group. The original d'Orbigny’s collection is kept in the National Museum of Natural History in Paris, and its copies were acquired by museums and universities in different countries. The copy of the d'Orbigny collection of the Paleontological Museum of St. Petersburg State University was probably made by the German trading company “Dr. F. Krantz, Rheinisches Mineralien-Kontor”. It was acquired by the St. Petersburg Imperial University for the Geological Cabinet (museum), which were created since 1870 by Professor A.A. Inostrantsev. Based on literary sources, the history of revisions of the d'Orbigny collection has been restored. That made it possible to correct errors in the museum labels, as well as to establish the original names of species depicted as models and bring them into line with the now accepted foraminiferal taxonomy. The second collection (No. ПЛ ОФ 75) is only a part (15 objects) of the collection of 100 models made by Václav Frič to illustrate foraminiferal classification scheme by the Austrian paleontologist August von Reuss. It was purchased for the Geological Cabinet of the Higher Women's (Bestuzhev) Courses and subsequently entered the Department of Paleontology, established at Petrograd University in 1919. As a result of the revision, the species affiliation of the species depicted as the models was established in accordance with the modern system of foraminifera.

Concurrent channel and spatial attention in Fully Convolutional Network for individual pig image segmentation

The separation of individual pigs from the pigpen scenes is crucial for precision farming, and the technology based on convolutional neural networks can provide a low-cost, non-contact, non-invasive method of pig image segmentation. However, two factors limit the development of this field. On the one hand, the individual pigs are easy to stick together, and the occlusion of debris such as pigpens can easily make the model misjudgment. On the other hand, manual labeling of group-raised pig data is time-consuming and labor-intensive and is prone to labeling errors. Therefore, it is urgent for an individual pig image segmentation model that can perform well in individual scenarios and can be easily migrated to a group-raised environment. In order to solve the above problems, taking individual pigs as research objects, an individual pig image segmentation dataset containing 2066 images was constructed, and a series of algorithms based on fully convolutional networks were proposed to solve the pig image segmentation problem. In order to capture the long-range dependencies and weaken the background information such as pigpens while enhancing the information of individual parts of pigs, the channel and spatial attention blocks were introduced into the best-performing decoders UNet and LinkNet. Experiments show that using ResNext50 as the encoder and Unet as the decoder as the basic model, adding two attention blocks at the same time achieves 98.30% and 96.71% on the F1 and IOU metrics, respectively. Compared with the model adding channel attention block alone, the two metrics are improved by 0.13% and 0.22%, respectively. The experiment of introducing channel and spatial attention alone shows that spatial attention is more effective than channel attention. Taking VGG16-LinkNet as an example, compared with channel attention, spatial attention improves the F1 and IOU metrics by 0.16% and 0.30%, respectively. Furthermore, the heatmap of the feature of different layers of the decoder after adding different attention information proves that with the increase of layers, the boundary of pig image segmentation is clearer. In order to verify the effectiveness of the individual pig image segmentation model in group-raised scenes, the transfer performance of the model is verified in three scenarios of high separation, deep adhesion, and pigpen occlusion. The experiments show that the segmentation results of adding attention information, especially the simultaneous fusion of channel and spatial attention blocks, are more refined and complete. The attention-based individual pig image segmentation model can be effectively transferred to the field of group-raised pigs and can provide a reference for its pre-segmentation. Keywords: pig, image segmentation, Fully Convolutional Network (FCN), attention mechanism, channel and spatial attention DOI: 10.25165/j.ijabe.20231601.6528 Citation: Hu Z W, Yang H, Lou T T, Yang H W. Concurrent channel and spatial attention in Fully Convolutional Network for individual pig image segmentation. Int J Agric & Biol Eng, 2023; 16(1): 232–242.

The nature of dispensing errors in hospital pharmacy; a systematic review

Medication errors can result in several unfavorable outcomes. Dispensing medication is the fundamental function of pharmaceutical care. Dispensing error is the discrepancy between the prescription and the drug delivered to the patients. The rates of dispensing errors vary in the literature across different countries and settings, and there are several forms and a wide range of dispensing errors as well as various causes. Hence, this review aimed to identify the prevalence, types, and causes of dispensing errors in hospital pharmacy by reviewing the previous studies that investigated this subject. Obtaining eligible articles required exploration of electronic databases, including PubMed, Scopus, Science Direct, Elsevier, Springer, and Google Scholar. The exploration process involved using search terms such as \"Nature, Prevalence, Incidence, Types, Causes, Dispensing errors, Hospital Pharmacy, Hospitals, and Rate\". The inclusion criteria were English original articles reported dispensing errors in hospital pharmacies published after the year 2016. A total of seven studies were included as they met the inclusion criteria. The studies were from six countries with a total number of prescriptions of 66,849. The nature of dispensing errors included prevalence rates, types, and causes of dispensing errors. Dispensing errors in hospital pharmacies are a significant issue that varies between different studies, even hospital settings. The major dispensing errors included labeling errors, content errors, and documentation errors. There were various factors that led to such dispensing errors, including a shift of work, heavy work, similar drug names and packaging, poor handwriting, interruption, a high number of drugs, and quantified drugs.

Synthetic Data Augmentation Using GAN For Improved Automated Visual Inspection

Quality control is a crucial activity manufacturing companies perform to ensure their products conform to the requirements and specifications. The introduction of artificial intelligence models enables to automate the visual quality inspection, speeding up the inspection process and ensuring all products are evaluated under the same criteria. In this research, we compare supervised and unsupervised defect detection techniques and explore data augmentation techniques to mitigate the data imbalance in the context of automated visual inspection. Furthermore, we use Generative Adversarial Networks for data augmentation to enhance the classifiers’ discriminative performance. Our results show that state-of-the-art unsupervised defect detection does not match the performance of supervised models but can reduce the labeling workload if tolerating some labeling errors. Furthermore, the best classification performance was achieved considering GAN-based data generation with AUC ROC scores equal to or higher than 0,9898. We performed the research with real-world data provided by Philips Consumer Lifestyle BV.

Arctic Sea Ice and Open Water Classification From Spaceborne Fully Polarimetric Synthetic Aperture Radar

We present an approach for automatic classification of Arctic sea ice and open water from spaceborne C-band RADARSAT-2 fully polarimetric (HH, HV, VH, VV) synthetic aperture radar (SAR) imagery, based on the random forest (RF) model. The HH- and HV-polarized radar backscatter, incidence angle and optimal polarimetric features are inputs of the RF model. <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">First</i> , we use a physics-based unsupervised scheme to generate well-annotated sea ice and open water samples. This scheme can alleviate labeling errors arising from visual interpretation or temporal-spatial mismatch between SAR images and operational ice charts. <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Subsequently</i> , we estimate a set of characteristic polarimetric parameters and select the most representative features, using the recursive feature elimination method. <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Finally</i> , the RF model is trained and validated using more than one million labelled samples. Statistical validation results show that the overall sea ice and water classification accuracy is 99.94% and the Kappa coefficient is 0.999. We find that ice–water discrimination accuracy can be improved by about 4%–10%, when optimal polarimetric features are involved in the RF model input. Moreover, the <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">polarization difference</i> (PD, VV–HH) is found to be the foremost polarimetric parameter for distinguishing sea ice from open water. The proposed approach has capability of yielding satisfactory ice–water classification over the complicated marginal ice zone. High-resolution ice–water classification maps clearly show that sea ice leads and their surroundings are also well separated.

A threat to patient safety due to drug labelling error in the postanaesthesia care unit during the COVID-19 pandemic

A threat to patient safety due to drug labelling error in the postanaesthesia care unit during the COVID-19 pandemic