Labeling Process Research Articles

Mechanical property prediction of random copolymers using uncertainty-based active learning

The copolymer, a widely used material in our daily lives, presents a significant challenge in targeted sequence design. While recent advancements in computational simulation and data science offer a promising avenue for addressing this complex issue, challenges persist in labeled data scarcity. In this study, we introduce an uncertainty-based active learning framework for predicting the properties of random copolymers. We found that the active learning strategy allowed for labeling only 40 data points within the design space of 1550 data points, drastically reducing the labeling efforts by 97%. Most data selected by active learning were positioned on the design space’s periphery, transforming the learning task into an interpolation problem. Through integrating active learning and molecular dynamics, we successfully overcame the combinatorial explosion problem in copolymer sequence design, streamlining the data labeling process and culminating in a highly accurate model. This research demonstrates data science’s potential in polymer design, especially when facing data scarcity.

Label-free miRNA fluorescent biosensors based on duplex-specific nucleases and silver nanoclusters.

MicroRNAs (miRNAs) are considered reliable biomarkers for a variety of diseases. However, their low abundance in organisms and high sequence similarity of homologous miRNAs make their accurate detection challenging. Here, we constructed a novel fluorescent biosensor for the detection of miRNA-155, a potential biomarker of neuroinflammation, based on duplex-specific nuclease (DSN) assisted amplification and DNA-templated silver nanoclusters (DNA-AgNCs) as fluorescence signal probes. DSN-assisted amplification can transform unstable miRNA into stable DNA and amplify the miRNA signal at the same time. Using DNA-AgNCs as fluorescence signal probes for biosensors can avoid complex labeling processes and reduce costs. The biosensor shows excellent selectivity, reproducibility, a wide linear range (1-600 nM) with a detection limit of 0.86 nM, and potentiality for real sample detection. This work provides a potential universal biosensing platform for miRNA detection.

Sentiment Analysis of National Health Security Mobile Application Review Using Machine Learning

It is widely acknowledged that digital transformation provides the opportunity for business process improvements. We can see many businesses, specifically in Indonesia's private and public sectors, leveraging technology to provide better service to their stakeholders. This research seeks insight into stakeholders’ engagement in digital transformation in the Indonesian healthcare system, namely the National Health Security Mobile application (Mobile JKN). This study employs a quantitative method to analyze user sentiment from Google Play reviews. Firstly, user reviews are extracted, preprocessing steps are applied, and machine learning-based sentiment labeling is employed afterward to categorize them into positive, neutral, and negative sentiments. The machine labeling process is carried out in document-level analysis, meaning user reviews represent their overall sentiment. Subsequently, we study the most prominent words using Word Cloud to determine the topics mainly discussed in positive and negative sentiments. The result is that 56.10% of reviews from 7 June 2016 to 14 July 2024 contain positive sentiments, while 43.17% contain negative sentiments. Neutral reviews contribute the most minor proportion, making up only 0.73%. The most prominent words in positive sentiment reviews, such as easy, sound, and helpful, suggest the user perception of the National Health Security Mobile as being easy to use and successfully accommodating the user's needs. In contrast, the words application, update, complex, login, code, verification, register, open, use, and error dominate the negative reviews, indicating users had difficulties logging in and registering user accounts, mainly related to frequent updates and Time Password errors.

RECOGNIZING A TRANSFORMER OIL GAUGE IN AN IMAGE USING A PRE-TRAINED MOBILENETV2 FPN LITE MODEL

Oil gauges allow you to monitor the amount of oil in transformers, which is necessary to prevent its overheating and related negative consequences, such as deterioration of the insulating and cooling properties of the oil, increase in pressure inside the device and possible accidents. Most of the oil pointers do not have an electric output to indicate the measurement results, which complicates the process of reading the readings of the device. One of the options for solving this problem may be the use of intelligent methods of image analysis, which have recently been actively used in many scientific and practical fields. The purpose of the study was to develop and evaluate the effectiveness of an intelligent system for determining the presence of an arrow transformer oil gauge in an image using a pre-trained MobileNetV2 FPN Lite model. Materials and methods. In the course of the study, theoretical and empirical methods were used. The study considers the pointer oil gauges of the MS-1 and MS-2 models. To develop a machine learning model for image recognition, the Python language was chosen, the open-source TensorFlow Object Detection API platform, which is part of TensorFlow, was used, which implements the creation, training, and testing of neural network object detection models. Research results. The process of generating a training set using augmentation methods and the process of data labeling are described. The process of selecting hyperparameters for model training was considered and analyzed. The hyperparameter “number of neurons” was used with the same value as in the pre-trained model. ReLU was chosen as the activation function, and the momentum optimization method was used. The batch size was 4, and the number of epochs was assumed to be 2000. The Accuracy, Recall, Precision, and F1-score metrics were calculated, and the quality and efficiency of the resulting model were evaluated based on them. The model did a good job of avoiding false positives, which confirms the high value of the Precision = 1.0 metric. However, it incorrectly recognized a significant number of non-oil gauge objects, as reflected by the low Recall = 0.687 value. The reason for such indicators of the Recall and F1-score metrics is the imbalance of data for training. To eliminate this drawback in further research, it is necessary to prepare a larger number of unique images of this device, displaying it at different angles of inclination, in different weather conditions, in different lighting. Conclusions. A model of an intelligent system for determining the presence of a transformer oil gauge in an image using a pre-trained MobileNetV2 FPN Lite model has been built. The model was trained to determine the device and its boundaries on the image, but the accuracy of the model is insufficient for its use in real conditions. Increasing the training, validation and test sample size can solve this problem. In general, the proposed method of recognizing transformer oil indicator on the image has shown its performance.

A Weakly Supervised Crowd Counting Method via Combining CNN and Transformer

During the past five years, there has been an increasing trend of weakly supervised crowd counting methods being developed since such methods just rely on count-level annotations and avoid a laborious labeling process. But, the existing weakly supervised methods usually fail to achieve comparable counting performance to the fully supervised methods. To improve the accuracy of crowd counting tasks, we propose to combine the convolutional neural network (CNN) and Transformer frameworks. Since CNN focuses on capturing local detail information and Transformer can effectively extract global context information, we believe that the combination of CNN and Transformer could learn more efficient feature representations for crowd images. Our proposed framework is named CrowdCCT (Crowd Counting via CNN and Transformer), and it is composed of a CNN feature extraction part, a Transformer feature extraction part, and a counting regression part. In the CNN part, we utilize DenseNet121 to learn rich semantic features with its inherent dense connection structure. In the Transformer part, we introduce two attention modules, Multi-Scale Dilated Attention (MSDA) and Location-Enhanced Attention (LEA), working together to extract more expressive features. The output features are then fed into the regression part to generate the predicted counting results. Experiments on four crowd counting benchmark datasets demonstrate that our proposed CrowdCCT can achieve superior performance. Also, the experimental results validate the feasibility and effectiveness of combining CNN and Transformer for weakly supervised counting tasks. Our work could be expected to promote further combination research on CNN and Transformer.

Classifying continuous GNSS stations using integrated machine learning

The development of Global Navigation Satellite Systems (GNSS) results in large spatial geodetic networks with a distinct range of accuracy. Thus, classification of the GNSS stations is needed to determine which stations are appropriate for geodetic applications. Additionally, advanced Machine Learning (ML) techniques have been proposed. However, ML algorithms may sometimes be less sensitive due to a lack of samples or anomalies in input data. Therefore, this study introduces an approach in which human-based supervision is integrated into ML processes to improve the ML model’s performance in classifying the continuous GNSS stations. The human factor influences the ML processes through two sampling strategies: “suggest-decide” and “correct-retrain”, where the accuracy of ML models will be improved via human-based corrections. The idea is that the unsupervised ML-based clustering techniques are driven by human-based supervision to create samples for training the supervised ML-based classification models. In this study, we develop a MATLAB app to automate the clustering and labeling processes. Our finding demonstrates that applying these sampling strategies can enhance the accuracy of the ML-based classification models from under 50 % up to ∼99 % after re-training. Also, this study categorizes almost 9000 continuous monitoring stations in the Nevada database, of which 1900 stations in Europe serve as samples for training the ML-based classification models. Furthermore, the methodologies developed in this study can be applied to warning systems, which utilize internal and external human resources to correct errors, address unusual situations, and provide timely feedback for better performance of ML-based forecasts.

Sentiment Analysis of Tiktok App Reviews on Google Play using Several Machine Learning Methods

Sentiment analysis has become increasingly important in understanding user perceptions of digital platforms. This study focuses on analyzing TikTok application reviews from the Google Play Store in Indonesia using machine learning techniques. The research aims to investigate sentiment distribution and compare the performance of three popular machine learning models: Random Forest, Support Vector Machine (SVM), and Naive Bayes. The study employed a comprehensive methodology involving data collection, preprocessing, feature extraction, and model evaluation. A dataset of 10,000 TikTok reviews was collected and preprocessed using techniques such as case folding, tokenization, and stopword removal. The sentiment labeling process categorizes reviews into positive, negative, and neutral sentiments based on user ratings. The TF-IDF algorithm was used for feature extraction, and the SMOTE technique addressed class imbalance. Results revealed a predominance of negative sentiment (53.5%), followed by neutral (32.1%) and positive (14.4%) sentiment. Model performance comparisons at different data sharing ratios (80/20 and 70/30) demonstrated that Random Forest and SVM consistently outperformed Naive Bayes. At the 80/20 ratio, Random Forest achieved the highest accuracy of 83.73%, highlighting its effectiveness in sentiment classification. The research contributes to the field of sentiment analysis and natural language processing by providing insights into user experiences with the TikTok application in Indonesia. The findings can guide application developers in understanding user perceptions and improving user experience.

Self Data Augmentation for Open Domain Question Answering

Information Retrieval (IR) constitutes a vital facet of Open-Domain Question Answering (ODQA) systems, focusing on the exploration of pertinent information within extensive collections of passages, such as Wikipedia, to facilitate subsequent reader processing. Historically, information retrieval relied on textual overlaps for relevant context retrieval, employing methods like BM25 and TF-IDF, which, however, lacked natural language understanding. The advent of deep learning ushered in a new era, leading to the introduction of Dense Passage Retrievers (DPR), shows superiority over traditional sparse retrievers. These dense retrievers leverage Pre-trained Language Models (PLMs) to initialize context encoders, enabling the extraction of natural language representations. They utilize the distance between latent vectors of contexts as a metric for assessing similarity. However, DPR methods are heavily reliant on large volumes of meticulously labeled data, such as Natural Questions. The process of data labeling is both costly and time-intensive. In this paper, we propose a novel data augmentation methodology SDA (Self Data Augmentation) that employs DPR models to automatically annotate unanswered questions. Specifically, we initiate the process by retrieving relevant pseudo passages for these unlabeled questions. We subsequently introduce three distinct passage selection methods to annotate these pseudo passages. Ultimately, we amalgamate the pseudo-labeled passages with the unanswered questions to create augmented data. Our experimental evaluations conducted on two extensive datasets (Natural Questions and TriviaQA), alongside a reletively small dataset (WebQuestions), utilizing three diverse base models, illustrate the significant enhancement achieved through the incorporation of freshly augmented data. Moreover, our proposed data augmentation method exhibits remarkable flexibility, which is readily adaptable to various dense retrievers. Additionally, we have conducted a comprehensive human study on the augmented data, which further supports our conclusions.

Overcoming the uncertainty challenges in detecting building changes from remote sensing images

Detecting building changes with multi-temporal remote sensing (RS) imagery at a very high resolution can help us understand urbanization and human activities, making informed decisions in urban planning, resource allocation, and infrastructure development. However, existing methods for building change detection (BCD) generally overlook critical uncertainty phenomena presented in RS imagery. Specifically, these uncertainties arise from two main sources: First, current building change detection datasets are designed primarily to detect changes in buildings, while changes in other land-cover classes are often classified as an unchanged background. Because of the manual labeling process, background elements that resemble buildings, such as roads and bridges, are at significant risk of being misclassified as building changes, introducing aleatoric uncertainty at the data level. Second, changes in parts of buildings that affect appearance, texture, or style without altering their semantic meaning, known as pseudo-changes, along with the imbalance between changed and unchanged samples, together lead to epistemic uncertainty at the model level. To address these challenges, we present an Uncertainty-Aware BCD (UA-BCD) framework. In detail, we employ a Siamese pyramid vision transformer to extract multi-level features from bi-temporal images, which are then decoded via a general decoder to obtain a coarse change map with inherent uncertainty. Subsequently, we introduce the aleatoric uncertainty estimation module to estimate the aleatoric uncertainty and embed it into the feature space. Then, a knowledge-guided feature enhancement module is developed to leverage the knowledge encoded in the coarse map to enhance the multi-level features and generate a refined change map. Finally, we propose an epistemic uncertainty estimator that takes the bi-temporal images and the refined change map as input and outputs an estimate of epistemic uncertainty. This estimation is supervised by the entropy calculated from the refined map, ensuring that the UA-BCD framework can produce a change map with lower epistemic uncertainty. To comprehensively validate the efficacy of the UA-BCD framework, we adopt a dual-perspective verification approach. Extensive experiments on five public building change datasets demonstrate the significant advantages of the proposed method over current state-of-the-art methods. Additionally, an application in Dongxihu District, Wuhan, China, confirms the outstanding performance of the proposed method in large-scale BCD. The source code of the project is available at https://github.com/Henryjiepanli/UA-BCD.

The Role of Packaging Designers in the Context of the Environmental Crisis

In the face of a worsening climate crisis and growing consumer awareness of the negative effects of over-consumption, the packaging industry faces the challenge of transforming itself towards more sustainable solutions. Labelling and well-planned production processes that can effectively promote environmental values that benefit both the producer and the environment play a crucial role in this process. Working with experts from other industries involved in creating goods, designers can shape informed and mature visual messages emphasizing the importance of responsible design decisions. The following paper emphasizes that sustainable design requires interdisciplinary knowledge and consideration of long-term effects, such as using and recycling materials, which can help reduce waste generation and promote responsible practices in the packaging industry.

HTAD: a human-in-the-loop framework for supervised chromatin domain detection

Topologically associating domains (TADs) are essential units of genome architecture, influencing transcriptional regulation and diseases. Despite numerous methods proposed for TAD identification, it remains challenging due to complex background and nested TAD structures. We introduce HTAD, a human-in-the-loop TAD caller that combines machine learning with human supervision to achieve high accuracy. HTAD begins with feature extraction for potential TAD border pairs, followed by an interactive labeling process through active learning. Performance assessments using public curation and synthetic datasets demonstrate HTAD’s superiority over other state-of-the-art methods and reveal highly hierarchical TAD structures, offering a human-in-the-loop solution for detecting complex genomic patterns.

Age- and sex-related differences in social competence and emotion labeling in pre-adolescence.

Identification of facial expressions is important to navigate social interactions and associates with developmental outcomes. It is presumed that social competence, behavioral emotion labeling and neural emotional face processing are related, but this has rarely been studied. Here, we investigated these interrelations and their associations with age and sex, in the YOUth cohort (1055 children, 8-11 years old). Using a multistep linear modelling approach, we associated parent-reported social competence, basic emotion labeling skills based on pictures of facial expressions, and neural facial emotion processing during a passive-watching fMRI task with pictures of houses and emotional faces. Results showed better emotion labeling and higher social competence for girls compared to boys. Age was positively associated with emotion labeling skills and specific social competence subscales. These age- and sex-differences were not reflected in brain function. During fMRI, happy faces elicited more activity than neutral or fearful faces. However, we did not find evidence for the hypothesized links between social competence and behavioral emotion labeling, and with neural activity. To conclude, in pre-adolescents, social competence and emotion labeling varied with age and sex, while social competence, emotion labeling and neural processing of emotional faces were not associated with each other.

Large-Scale Object Detection in the Wild With Imbalanced Data Distribution, and Multi-Labels.

Training with more data has always been the most stable and effective way of improving performance in the deep learning era. The Open Images dataset, the largest object detection dataset, presents significant opportunities and challenges for general and sophisticated scenarios. However, its semi-automatic collection and labeling process, designed to manage the huge data scale, leads to label-related problems, including explicit or implicit multiple labels per object and highly imbalanced label distribution. In this work, we quantitatively analyze the major problems in large-scale object detection and provide a detailed yet comprehensive demonstration of our solutions. First, we design a concurrent softmax to handle the multi-label problems in object detection and propose a soft-balance sampling method with a hybrid training scheduler to address the label imbalance. This approach yields a notable improvement of 3.34 points, achieving the best single-model performance with a mAP of 60.90% on the public object detection test set of Open Images. Then, we introduce a well-designed ensemble mechanism that substantially enhances the performance of the single model, achieving an overall mAP of 67.17%, which is 4.29 points higher than the best result from the Open Images public test 2018.

SEGMENTASI LESI KULIT MONKEYPOX MENGGUNAKAN ARSITEKTUR U-NET

Monkeypox has become an issue of global concern as cases continue to rise across various countries. While most cases display mild symptoms, severe infections can lead to life-threatening complications. Early detection is therefore crucial to control the spread of monkeypox, with one approach being the segmentation of monkeypox lesions to differentiate affected areas from healthy skin and remove any interfering elements, such as hair, which can enhance model accuracy. In this consider, due to the inaccessibility of labeled masks, mask information were produced utilizing watershed segmentation with the sobel operator and otsu thresholding. The division of skin injuries was in this way carried out with a U-Net demonstrate, utilizing MobileNetV2 as the backbone and ImageNet weights for transfer learning. The U-Net model reached an accuracy of 88.07%, though some signs of overfitting were observed, likely due to low-quality label information from the watershed labeling process, which necessitates parameter tuning.

Detection of Critical Parts of River Crab Based on Lightweight YOLOv7-SPSD.

The removal of back armor marks the first stage in the comprehensive processing of river crabs. However, the current low level of mechanization undermines the effectiveness of this process. By integrating robotic systems with image recognition technology, the efficient removal of dorsal armor from river crabs is anticipated. This approach relies on the rapid identification and precise positioning of the processing location at the crab's tail, both of which are essential for optimal results. In this paper, we propose a lightweight deep learning model called YOLOv7-SPSD for detecting river crab tails. The goal is to accurately determine the processing location for the robotic removal of river crab back armor. We start by constructing a crab tail dataset and completing the data labeling process. To enhance the lightweight nature of the YOLOv7-tiny model, we incorporate the Slimneck module, PConv, and the SimAM attention mechanism. These additions help achieve an initial reduction in model size while preserving detection accuracy. Furthermore, we optimize the model by removing redundant parameters using the DepGraph pruning algorithm, which facilitates its application on edge devices. Experimental results show that the lightweight YOLOv7-SPSD model achieves a mean Average Precision (mAP) of 99.6% at a threshold of 0.5, an F1-score of 99.6%, and processes frames at a rate of 7.1 frames per second (FPS) on a CPU. Compared to YOLOv7-tiny, the improved model increases FPS by 2.7, reduces GFLOPS by 74.6%, decreases the number of parameters by 71.6%, and lowers its size by 8.1 MB. This study enhances the deployment of models in river crab processing equipment and introduces innovative concepts and methodologies for advancing intelligent river crab deep processing technology.

Comparison of Naive Bayes and SVM Algorithms for Sentiment Analysis of PUBG Mobile on Google Play Store

PlayerUnknown's Battlegrounds (PUBG) Mobile is one of the most popular mobile games in Indonesia, according to data from the Google Play Store. According to the Google Play Store, the game has a rating of 3.8 with 49.5 million reviews. While a considerable number of users express satisfaction, a significant proportion of reviews also contain criticism regarding the gameplay and features. However, a cursory examination of reviews may not fully capture the nuances of user sentiment, necessitating a more comprehensive sentiment analysis. This research will employ a positive and negative sentiment analysis of Indonesian PUBG Mobile reviews on the Google Play Store, utilizing a comparative approach to evaluate the performance of two algorithms: Naïve Bayes and Support Vector Machine (SVM). The data set comprised 2,000 user reviews, which were collected using a scraping technique. Following this, a labeling process was conducted based on the rating, data were preprocessed, TF-IDF weighting was applied, and both algorithms were implemented. The findings indicated that users expressed satisfaction with the game's visuals and gameplay. However, there were also technical concerns that required attention, including bugs, server instability, lag, and performance issues. The SVM algorithm demonstrated superior performance, with an accuracy rate of 70.95%, compared to Naïve Bayes, which reached 69.83%. Despite Naïve Bayes's faster processing speed, SVM exhibited greater precision, recall, and F1-score

CASSAD: Chroma-Augmented Semi-Supervised Anomaly Detection for Conveyor Belt Idlers.

Idlers are essential to conveyor systems, as well as supporting and guiding belts to ensure production efficiency. Proper idler maintenance prevents failures, reduces downtime, cuts costs, and improves reliability. Most studies on idler fault detection rely on supervised methods, which depend on large labelled datasets for training. However, acquiring such labelled data is often challenging in industrial environments due to the rarity of faults and the labour-intensive nature of the labelling process. To address this, we propose the chroma-augmented semi-supervised anomaly detection (CASSAD) method, designed to perform effectively with limited labelled data. At the core of CASSAD is the one-class SVM (OC-SVM), a model specifically developed for anomaly detection in cases where labelled anomalies are scarce. We also compare CASSAD's performance with other common models like the local outlier factor (LOF) and isolation forest (iForest), evaluating each with the area under the curve (AUC) to assess their ability to distinguish between normal and anomalous data. CASSAD introduces chroma features, such as chroma energy normalised statistics (CENS), the constant-Q transform (CQT), and the chroma short-time Fourier transform (STFT), enhanced through filtering to capture rich harmonic information from idler sounds. To reduce feature complexity, we utilize the mean and standard deviation (std) across chroma features. The dataset is further augmented using additive white Gaussian noise (AWGN). Testing on an industrial dataset of idler sounds, CASSAD achieved an AUC of 96% and an accuracy of 91%, surpassing a baseline autoencoder and other traditional models. These results demonstrate the model's robustness in detecting anomalies with minimal dependence on labelled data, offering a practical solution for industries with limited labelled datasets.

Enhancing Stock Market Prediction with Sentiment Analysis Using a BERT-based Model

This study explores the prediction of stock price trends in the financial market, emphasizing the impact of investor sentiment and macroeconomic policies. Traditional research often uses mathematical, statistical, or deep learning methods to predict stock prices but overlooks the emotional factors in vast unstructured text data, such as financial news. This paper proposes a Bidirectional Encoder Representations from Transformers (BERT)-Transformer model that integrates sentiment analysis to enhance stock market prediction. Using financial news text data from the Oriental Fortune Network, the BERT model performs sentiment analysis to extract emotional polarity features. These features, combined with stock transaction data, are then input into a Transformer model to predict the index trends. The experimental results demonstrate a 60% accuracy in predicting stock indices' rise and fall, indicating the model's effectiveness while highlighting areas for improvement. The methodology details dataset preprocessing, the BERT- Back Propagation (BP) model structure, and how sentiment classification is combined with stock trading data for predictions. Performance comparisons with other prediction models confirm the benefits of incorporating sentiment analysis. The BERT-Transformer model’s long-term memory capabilities make it a promising tool for financial time series predictions, offering new research ideas and methods for the financial field. Future research will aim to automate the sentiment feature labeling process to save time and improve efficiency.

Aptamer Immobilization on Transistor-Based Biosensor Via Crosslinker for Amine-to-Thiol Conjugation Toward Cortisol Detection

A Field-effect transistor (FET) biosensor is a sensing device that detect changes in the interfacial potential associated with the adsorption of target molecules. FET biosensor is expected to be a simple measurement method by detecting target biomarkers without any labeling process. Recently, aptamers, single-stranded DNA molecules with specificity and stability, have attracted attention as receptors of FET biosensor [1,2]. Immobilization of aptamer onto metal oxide insulator surface of FET biosensors is often achieved by combining self-assembled monolayer (SAM) with cross-linking agents. Generally, an immobilization method using cross-linking agent, glutaraldehyde, between amino-modified aptamer and aminopropylsilane (APS) monolayer is often employed [3]. However, there is concern that aptamers may be immobilized by non-terminal amino group, because amino groups are also present within the bases constituting DNA structure. Consequently, decrease in affinity of the aptamer to the target molecule by involving collapse of the higher-order structure and inhibition of conformational change. In this study, we attempted immobilization method on the FET gate surface utilizing cross-linking thiol groups of aptamers to self-assembled monolayer instead of that between amino groups. By immobilizing thiol-modified aptamers onto APS monolayer via N-succinimidyl 3-(2-pyridyldithio) propionate (SPDP), we demonstrated specific immobilization at the aptamer termini for increasing the number of effective aptamers for binding of cortisol, which is stress-related hormone, was selected as a target.The SiO2 gate insulator of the FET was treated by piranha solution (H2O2 : H2SO4 = 1:4) for introduction of hydroxyl groups. Then, the FET chip was immersed in toluene solvent including 1%(v/w) 3-aminopropyltriethoxysilane in an argon atmosphere (60ºC for 7 min.). The SPDP was reacted with amino group of APS monolayer. After that, dithiothreitol (DTT) was added to the FET gate surface to form thiol group by cleavage of the disulfide bond within the SPDP. 100 nM 3’-thiol modified cortisol aptamer was immobilized to the FET gate surface by forming disulfide bond. Finally, 1 μM 2-mercaptoethanol was reacted with residual thiol sites to prevent non-specific adsorption of contaminating proteins. The FET characteristics were measured before and after the addition of cortisol for obtaining the threshold voltage shift (ΔV g).First, we compared the ability of reducing regent for cleavage of disulfide bond within SPDP to form the thiol group on the FET gate surface. A 5,5'-dithiobis(2-nitrobenzoic acid) (DTNB), which reacts with thiol group, was added to SPDP-modified surface treated by DTT, ascorbic acid or l-cysteine ethyl ester (LCEE). Following that, the absorbance derived from 5-mercapto-2-nitrobenzoic acid, which was generated from DTNB depending on the amount of thiol groups, was measured. As a result, thiol groups derived from SPDP were formed on the FET gate surface upon treatment with DTT. On the other hand, no absorbance of 5-mercapto-2-nitrobenzoic acid was observed when ascorbic acid or LCEE was used as reducing regents. Additionally, thiol groups were increased with SPDP concentrations in the range from 0.5 to 25 mM by DTNB absorbance measurements. Subsequently, electrical responses to cortisol using aptamer-immobilized FET were measured upon varying SPDP concentrations, with maximum response obtained at 1 mM. It should be noted that the FET response was acquired by increase in charge within the Debye length due to the conformational changes of the aptamer upon cortisol binding [4]. Finally, the sensitivity of aptamer-immobilized FET biosensor, which was fabricated using SPDP, were compared with that obtained using glutaraldehyde as the cross-linking agent. As a result, the gate voltage shift by addition of 1 μM cortisol when SPDP was utilized to fabricate the aptamer-immobilized FET biosensor was greater than the case of glutaraldehyde (Figure 1). These results indicated that cortisol was efficiently captured by the increase in the amount of aptamer, which was immobilized by 3’-terminus via disulfide bond. Therefore, the utilization of cross-linker for amine-to-thiol conjugation to nucleic acid immobilization could be useful for improving the sensitivity of FET biosensors.[1] H. Chen et al., Analyst, 141, 2335-2346 (2016).[2] N. Nakatsuka et al., Science, 362, 319-324 (2018).[3] S. K. Vashist et al., Chem. Rev., 114, 11083−11130 (2014).[4] H. Hayashi et al., Electrochemistry, 89, (2), 134-137 (2021). Figure 1

An Efficient Printing Defect Detection Based on YOLOv5-DCN-LSK.

During the production process of inkjet printing labels, printing defects can occur, affecting the readability of product information. The distinctive shapes and subtlety of printing defects present a significant challenge for achieving high accuracy and rapid detection in existing deep learning-based defect detection systems. To overcome this problem, we propose an improved model based on the structure of the YOLOv5 network to enhance the detection performance of printing defects. The main improvements include the following: First, we introduce the C3-DCN module to replace the C3 module in the backbone network, enhancing the model's ability to detect narrow and elongated defects. Secondly, we incorporate the Large Selective Kernel (LSK) and RepConv modules into the feature fusion network, while also integrating a loss function that combines Normalized Gaussian Wasserstein Distance (NWD) with Efficient IoU (EIoU) to enhance the model's focus on small targets. Finally, we apply model pruning techniques to reduce the model's size and parameter count, thereby achieving faster detection. Experimental results demonstrate that the improved YOLOv5 achieved a mAP@0.5 of 0.741 after training, with 323.2 FPS, which is 2.7 and 20.8% higher than that of YOLOv5, respectively. The method meets the requirements of high precision and high efficiency for printing defect detection.