Transfer Learning Research Articles

A deep learning method for comparing Bayesian hierarchical models.

Bayesian model comparison (BMC) offers a principled approach to assessing the relative merits of competing computational models and propagating uncertainty into model selection decisions. However, BMC is often intractable for the popular class of hierarchical models due to their high-dimensional nested parameter structure. To address this intractability, we propose a deep learning method for performing BMC on any set of hierarchical models which can be instantiated as probabilistic programs. Since our method enables amortized inference, it allows efficient re-estimation of posterior model probabilities and fast performance validation prior to any real-data application. In a series of extensive validation studies, we benchmark the performance of our method against the state-of-the-art bridge sampling method and demonstrate excellent amortized inference across all BMC settings. We then showcase our method by comparing four hierarchical evidence accumulation models that have previously been deemed intractable for BMC due to partly implicit likelihoods. Additionally, we demonstrate how transfer learning can be leveraged to enhance training efficiency. We provide reproducible code for all analyses and an open-source implementation of our method. (PsycInfo Database Record (c) 2024 APA, all rights reserved).

GPS-SUMO 2.0: an updated online service for the prediction of SUMOylation sites and SUMO-interacting motifs.

Small ubiquitin-like modifiers (SUMOs) are tiny but important protein regulators involved in orchestrating a broad spectrum of biological processes, either by covalently modifying protein substrates or by noncovalently interacting with other proteins. Here, we report an updated server, GPS-SUMO 2.0, for the prediction of SUMOylation sites and SUMO-interacting motifs (SIMs). For predictor training, we adopted three machine learning algorithms, penalized logistic regression (PLR), a deep neural network (DNN), and a transformer, and used 52404 nonredundant SUMOylation sites in 8262 proteins and 163 SIMs in 102 proteins. To further increase the accuracy of predicting SUMOylation sites, a pretraining model was first constructed using 145545 protein lysine modification sites, followed by transfer learning to fine-tune the model. GPS-SUMO 2.0 exhibited greater accuracy in predicting SUMOylation sites than did other existing tools. For users, one or multiple protein sequences or identifiers can be input, and the prediction results are shown in a tabular list. In addition to the basic statistics, we integrated knowledge from 35 public resources to annotate SUMOylation sites or SIMs. The GPS-SUMO 2.0 server is freely available at https://sumo.biocuckoo.cn/. We believe that GPS-SUMO 2.0 can serve as a useful tool for further analysis of SUMOylation and SUMO interactions.

Hybrid Transfer Learning and Support Vector Machine Models for Asphalt Pavement Distress Classification

Pavement condition evaluation plays a crucial role in assisting with the management of the highway infrastructure. However, the current methods used for assessing pavement conditions are costly, time-consuming, and subjective. There is a growing need to automate these assessment tactics and leverage low-cost technologies to enable widespread deployment. This study aims to develop robust and highly accurate models for classifying asphalt pavement distresses using transfer learning (TL) techniques based on pretrained deep learning (DL) networks. This topic has gained considerable attention in the field since 2015 when DL became the mainstream choice for various computer vision tasks. While progress has been made in TL model development, challenges persist in areas of accuracy, repeatability, and training cost. To tackle these challenges, this study proposes hybrid models that combine DL networks with support vector machines (SVMs). Three strategies were evaluated: single DL models using transfer learning (TLDL), hybrid models combining DL and SVM (DL+SVM), and hybrid models combining TLDL and SVM (TLDL+SVM). The performance of each strategy was assessed using statistical metrics based on the confusion matrix. Results consistently showed that the TLDL+SVM strategy outperformed the other approaches in accuracy and F1 scores, regardless of the DL network type. On average, the hybrid models achieved an accuracy of 95%, surpassing the 80% accuracy of the best single model and the 55% accuracy for DL+SVM without TL. The results clearly indicate that employing transfer-learned models as feature extractors, in combination with SVM as the classifier, consistently achieves exceptional performance.

Chest Xray Medical Diagnosis with Deep Learning

Detecting Disease in Chest X-Rays with Deep Learning Chest X-rays (CXRs) is a common and vital diagnostic tool, but interpreting them accurately requires trained radiologists who can be scarce in many areas. This research explores the potential of deep learning (DL), a type of artificial intelligence (AI), to automatically detect diseases in CXRs. The review explains the basics of DL for CXR analysis, including how deep neural networks (DNNs) work and how transfer learning and data augmentation techniques can improve their performance. It then examines recent studies on applying DNNs to identify common CXR abnormalities like lung nodules, pneumonia, and pneumothorax. This includes exploring multi-class classification, where the model can detect multiple diseases simultaneously. The abstract compares the performance of these techniques with human observers and discusses the challenges of implementing DNN models in clinical practice, particularly their relationship with radiologists. Overall, the research investigates how DL can potentially improve disease detection using CXRs, addressing the need for accurate and accessible diagnostic tools. Key Words: Chest X-rays, Deep Learning, Disease Detection, Artificial Intelligence, Neural Networks

A machine learning approach for package size estimation using UHF RFID interrogation signature

AbstractThis paper introduces a new approach for performing package classification and sizing using Radio-Frequency Identification (RFID) systems. This technique is applicable when packages are labeled with or contain multiple RFID-tagged items. During the interrogation of the tags, received signal strength (RSS) statistics and other information, such as the frame count or the reading time, are collected by the reader and used to predict the package type from a set of candidate classes using an Artificial Neural Network (ANN). The primary challenge lies in acquiring sufficient training data for a target scenario to ensure reliable predictions. To address this, a two-phase training process based on transfer learning is adopted. Initially, a base model is developed using synthetic data generated from a detailed RFID simulator, designed to suit diverse scenarios, establish detailed link budgets, and comprehensively simulate the communication protocols. This model is then refined using a small dataset collected experimentally in the actual scenario. This method was validated in a real testbed with four different package types. The base model was trained using 1000 synthetic samples per package type (4000 in total), whereas the refined model was trained with a dataset consisting of only 25 real interrogation traces (samples) per package type (100 in total). The experimental samples were obtained using a software-defined radio unit, the Ettus B210 Universal Software Radio Peripheral (USRP) platform. This experiment achieved an accuracy of over 92%. In summary, this approach introduces a new feature to existing RFID setups, demonstrating potential for advanced package handling and cost optimization in the logistics sector.

Evaluating Deep Learning Techniques for Detecting Aneurysmal Subarachnoid Hemorrhage: A Comparative Analysis of Convolutional Neural Network and Transfer Learning Models.

Machine learning (ML) and deep learning (DL) techniques offer a promising multidisciplinary solution for subarachnoid hemorrhage (SAH) detection. The novel transfer learning approach mitigates the time constraints associated with the traditional techniques and demonstrates a superior performance. This study aims to evaluate the effectiveness of convolutional neural networks (CNNs) and CNN-based transfer learning models in differentiating between aneurysmal subarachnoid hemorrhage (aSAH) and nonaneurysmal subarachnoid hemorrhage (naSAH). Data from Istanbul Ümraniye Training and Research Hospital, which included 15,600 DICOM images from 123 patients with aSAH and 7,793 images from 80 patients with naSAH, were used. The study employed four models: Inception-V3, EfficientNetB4, single-layer CNN, and three-layer CNN. Transfer learning models were customized by modifying the last three layers and using the Adam optimizer. The models were trained on Google Collaboratory and evaluated based on metrics such as F-score, precision, recall, and accuracy. EfficientNetB4 demonstrated the highest accuracy (99.92%), with a better F-score (99.82%), recall (99.92%), and precision (99.90%) than the other methods. The single- and three-layer CNNs and the transfer learning models produced comparable results. No overfitting was observed, and robust models were developed. CNN-based transfer learning models can accurately diagnose the etiology of SAH from CT images and is a valuable tool for clinicians. This approach could reduce the need for invasive procedures such as digital subtraction angiography, leading to more efficient medical resource utilization and improved patient outcomes.

A novel nondestructive detection approach for seed cotton lint percentage using deep learning

BackgroundThe lint percentage of seed cotton is one of the most important parameters for evaluating seed cotton quality and affects its price. The traditional measuring method of lint percentage is labor-intensive and time-consuming; thus, an efficient and accurate measurement method is needed. In recent years, classification-based deep learning and computer vision have shown promise in solving various classification tasks.ResultsIn this study, we propose a new approach for detecting the lint percentage using MobileNetV2 and transfer learning. The model is deployed on a lint percentage detection instrument, which can rapidly and accurately determine the lint percentage of seed cotton. We evaluated the performance of the proposed approach using a dataset comprising 66 924 seed cotton images from different regions of China. The results of the experiments showed that the model with transfer learning achieved an average classification accuracy of 98.43%, with an average precision of 94.97%, an average recall of 95.26%, and an average F1-score of 95.20%. Furthermore, the proposed classification model achieved an average accuracy of 97.22% in calculating the lint percentage, showing no significant difference from the performance of experts (independent-sample t-test, t = 0.019, P = 0.860).ConclusionThis study demonstrated the effectiveness of the MobileNetV2 model and transfer learning in calculating the lint percentage of seed cotton. The proposed approach is a promising alternative to traditional methods, providing a rapid and accurate solution for the industry.

Integrated End-to-End automatic speech recognition for languages for agglutinative languages

The relevance of the problem of automatic speech recognition lies in the lack of research for low-resource languages, stemming from limited training data and the necessity for new technologies to enhance efficiency and performance. The purpose of this work was to study the main aspects of integrated end-to-end speech recognition and the use of modern technologies in the natural processing of agglutinative languages, including Kazakh. In this article, the study of language models was carried out using comparative, graphic, statistical and analytical-synthetic methods, which were used in combination. This paper addresses automatic speech recognition (ASR) in agglutinative languages, particularly Kazakh, through a unified neural network model that integrates both acoustic and language modeling. Employing advanced techniques like connectionist temporal classification and attention mechanisms, the study focuses on effective speech-to-text transcription for languages with complex morphologies. Transfer learning from high-resource languages helps mitigate data scarcity in languages such as Kazakh, Kyrgyz, Uzbek, Turkish, and Azerbaijani. The research assesses model performance, underscores ASR challenges, and proposes advancements for these languages. It includes a comparative analysis of phonetic and word-formation features in agglutinative Turkic languages, using statistical data. The findings aid further research in linguistics and technology for enhancing speech recognition and synthesis, contributing to voice identification and automation processes.

A Comparative Dataset of Annotated Broccoli Heads Recorded with Depth Cameras from a Moving Vehicle

An extensive, publicly available dataset is presented—the LAR Broccoli dataset—which contains 20,000 manually annotated images of broccoli heads captured from a moving tractor at an organic farm in the UK. The dataset contains images of the same row of broccoli heads recorded at 30 frames per second (fps) with three different cameras. Two off-the-shelf, relatively low-cost depth-sensing cameras were used, with the tractor moving at a speed of around 1 km/h, in addition to a webcam, with the tractor moving twice as fast. The utility of the dataset is demonstrated in four ways. First, three different state-of-the-art detector models were trained on the dataset, achieving an overall mean Average Precision (mAP) score of over 95% for the best-performing detector. The results validate the utility of the dataset for the standard task of in-field broccoli head recognition. Second, experiments with transfer learning were conducted, initialised with a smaller pre-trained broccoli detection model, and refined with the LAR Broccoli dataset. Third, we assessed the advantages of transfer learning not only using mAP but also according to time and space requirements for training models, which provides a proxy metric for energy efficiency, a practical consideration for real-world model training. Fourth, the cross-camera generalisation among the three camera systems was compared. The results highlight that testing and training detector models using different camera systems can lead to reduced performance, unless the training set also includes some images captured in the same manner as those in the test set.

Transformer based ensemble deep learning approach for remote sensing natural scene classification

ABSTRACT Very high resolution (VHR) remote sensing (RS) image classification is paramount for detailed Earth’s surface analysis. Feature extraction from VHR natural scenes is crucial, but it becomes a challenging task because of the overlapping edges present in images. Multiple open-source datasets exist in the literature to train robust models, and they have been benchmarked using deep learning models. However, different datasets contain different numbers of classes; a few of them could be absent in other datasets because they are independent of the other classes, and the classes are sometimes not mutually exclusive amongst the same dataset. Thus, it is very challenging to generalize a model trained on a single dataset to perform scene classification on unknown classes of multiple benchmark datasets and real-time images. Thus, this work introduces the Remote Sensing Natural Scenes 92 (RS_NS92) dataset, consisting of 36,785 images belonging to 92 classes, curated by selectively taking the union of all subclasses from five benchmark datasets. This class count is significantly higher than publicly available datasets and maintains a low-class imbalance and a comprehensive data distribution for robust model training. It also provides the remote sensing community with an extra platform to validate the performance on multiple benchmarks. Inspired by federated learning, an ensemble approach consisting of three feature extraction backbones: Vision Transformers, Swin Transformers, and ConvNeXt (termed the VSC_Ensemble model) is also introduced. This model can make extraordinary predictions across multiple datasets by finetuning weights using transfer learning. Experimental analysis with the proposed approach not only obtains a high test accuracy of 97.24% and an F1-score of 0.9587 for the 92 classes on a 90:10 split of the proposed benchmark dataset but also gets excellent results on unseen test images of other datasets, which are comparable to the state-of-the-art results.

MTTLA‐DLW: Multi‐task TCN‐Bi‐LSTM transfer learning approach with dynamic loss weights based on feature correlations of the training samples for short‐term wind power prediction

AbstractWind power prediction for newly built wind farms is usually faced with the problem of no sufficient historical data. To efficiently extract the useful features from related wind farms, a novel transfer learning method based on temporal convolutional network (TCN)‐Bi‐long short‐term memory (LSTM) with dynamic loss weights is proposed. Firstly, a novel multi‐task TCN‐Bi‐LSTM model is designed to extract common features. The separate TCNs, and common Bi‐LSTM layers of the proposed model are designed to extract the temporal features from related wind farms. Secondly, in the pre‐training stage, to optimize the training process of the neural networks, a dynamic loss‐weighting strategy is proposed for multi‐task learning (MTL) to select the most related features, which increase the prediction accuracy by providing a suitable optimization object. Thirdly, the multi‐task TCN‐Bi‐LSTM model is re‐trained based on the samples from the target wind farm. Finally, a dataset of seven wind farms was employed to evaluate the efficiency of the proposed MTL structure and the dynamic loss‐weighting strategy. The result shows that the root mean squared error of the 12‐h short‐term prediction can be decreased by 4.19% compared with the traditional single‐task learning model, which verifies the validity of the proposed multi‐task transfer learning method.

Online detection of the incidence via transfer learning

AbstractThe counting process has abundant applications in reality, and Poisson process monitoring actually has received extensive attention and research. However, conventional methods experience poor performance when shifts appears early and only small number of historical observations in Phase I can be used for estimation. To overcome it, we creatively propose a new online monitoring algorithm under the transfer learning framework, which utilizes the information from observations of additional data sources so that the target process can be described better. By making the utmost of the somewhat correlated data from other domains, which is measured by a bivariate Gamma distributed statistic presented by us, the explicit properties (e.g., posterior probability mass function, posterior expectation, and posterior variance) are also strictly proved. Furthermore, based on the above theoretical results, we design two computationally efficient control schemes in Phase II, that is a control chart based on the cumulative distribution function for large shifts and an exponentially weighted moving average control chart for small shifts. For a better understanding of the more practical applications and transferability matter, we provide some optimal values for parameter setting. Extensive numerical simulations and a case of skin cancer incidence in America verify the superiorities of our approach.

A lateralized motor network in order to understand adaptation to visuomotor rotation

Objective. The functional asymmetry between the two brain hemispheres in language and spatial processing is well documented. However, a description of difference in control between the two hemispheres in motor function is not well established. Our primary objective in this study was to examine the distribution of control in the motor hierarchy and its variation across hemispheres. Approach. We developed a computation model termed the bilateral control network and implemented the same in a neural network framework to be used to replicate certain experimental results. The network consists of a simple arm model capable of making movements in 2D space and a motor hierarchy with separate elements coding target location, estimated position of arm, direction, and distance to be moved by the arm, and the motor command sent to the arm. The main assumption made here is the division of direction and distance coding between the two hemispheres with distance coded in the non-dominant and direction coded in the dominant hemisphere. Main results. With this assumption, the network was able to show main results observed in visuomotor adaptation studies. Importantly it showed decrease in error exhibited by the untrained arm while the other arm underwent training compared to the corresponding naïve arm’s performance—transfer of motor learning from trained to the untrained arm. It also showed how this varied depending on the performance variable used—with distance as the measure, the non-dominant arm showed transfer and with direction, dominant arm showed transfer. Significance. Our results indicate the possibility of shared control between the two hemispheres. If indeed found true, this result could have major significance in motor rehabilitation as treatment strategies will need to be designed in order to account for this and can no longer be confined to the arm contralateral to the affected hemisphere.

Genetic algorithm based deep learning model adaptation for improvising the motor imagery classification

ABSTRACT Deep learning methods have proved a promising performance for electroencephalography-based brain-computer interfaces (EEG-BCI). It is particularly encouraging that a subject-independent model can be trained using a large amount of other subjects’ data. Transfer learning methods such as adaptation or fine-tuning can be used on the pre-trained model to improve the performance. This study examined the influence of fine-tuning on the subject-independent model for EEG-based motor imagery (MI) classification using a genetic algorithm (GA). The proposed method is evaluated on the binary class MI dataset from the Korea University EEG dataset. Results show that the proposed GA-based fine-tuning approach statistically improved the average classification accuracy of the baseline model from 84.46% to 87.29%. More interestingly, our approach shows significant improvement in cases where the performance of the baseline model is poor after fine-tuning using other approaches. Further, layer-wise relevance propagation (LRP) is used to analyze the adapted models to gain a deeper understanding of the neurophysiological explanations underlying the model’s decision.

Deep learning based fault detection of automobile dry clutch system using spectrogram plots

Abstract In this research study, a novel approach that leverages the power of transfer learning (a famous deep learning technique) is proposed to diagnose multiple types of clutch faults including, worn release fingers, fractured pressure plates, deteriorated pressure plates, loss of friction material and distorted tangential strips using spectrogram plots. To train and validate the diagnostic system, vibration readings were taken from a specially designed test rig with the help of piezoelectric accelerometer while the clutch system was operated under different load conditions of 0 (no load), 5 and 10 kg This procedure of data collection was then repeated to acquire the vibration data for all of the fault conditions by replacing the good with fault components individually. These vibration signals were further processed and transformed into spectrogram plots that serves as the input data for the deep learning models considered. Fine-tuning techniques were applied on pretrained networks to maximise the prediction accuracy of the models to effectively determine and diagnose faults in the clutch system. For this study 12 pre-trained networks were chosen namely, Xception, InceptionResNet, DenseNet, AlexNet, VGG16, GoogLeNet, VGG19, ResNet101, ResNet50, InceptionV3, MobileNetV2 and ShuffleNet. To optimize the performance of deep learning models, a systematic adjustment of hyperparameters such as the train-test split ratio, learning rate, optimizer and batch size for each network model was carried out. Through careful experimentation and analysis, significant improvements in fault classification accuracy were achieved thereby enhancing the reliability and effectiveness of the diagnostic system. From the results it was noted that 100% classification accuracy was displayed by AlexNet (for the no load condition and the 10kg load condition) and GoogLeNet (for 5 kg load condition) with extremely low computation times.

Demonstrating computer vision to small‐ and medium‐sized enterprises in manufacturing: Toward overcoming costs and implementation challenges

AbstractComputer vision (CV) systems are crucial in various fields, such as manufacturing. However, the general feedback from small and medium‐sized enterprises (SMEs) in manufacturing indicates that implementing CV systems are out of reach due to their high costs, the need for expertise, and the complexities of Machine Learning (ML). To demonstrate the feasibility and applicability for SMEs, we present a cost‐effective portable CV‐based demonstrator for SMEs in manufacturing. Incorporating low‐cost hardware components, the system integrates open‐source software and employs Transfer Learning to adapt pre‐existing ML models. We present two illustrative use cases with fault detection and inventory management based on plastic brick datasets including various colors and shapes to demonstrate the system's effectiveness. Bringing this into a showable demonstrator, all components are integrated into a portable suitcase as a plug and play demonstration. We showcase the portable demonstrator at German television and four different industrial fairs, leveraging these dynamic platforms for direct interaction with SMEs and stakeholders. Firsthand insights and feedback from SMEs regarding our demonstration and their challenges, as well as opportunities for CV in manufacturing were received and summarized in this research.

Apriori prediction of chemotherapy response in locally advanced breast cancer patients using CT imaging and deep learning: transformer versus transfer learning

ObjectiveNeoadjuvant chemotherapy (NAC) is a key element of treatment for locally advanced breast cancer (LABC). Predicting the response to NAC for patients with Locally Advanced Breast Cancer (LABC) before treatment initiation could be beneficial to optimize therapy, ensuring the administration of effective treatments. The objective of the work here was to develop a predictive model to predict tumor response to NAC for LABC using deep learning networks and computed tomography (CT).Materials and methodsSeveral deep learning approaches were investigated including ViT transformer and VGG16, VGG19, ResNet-50, Res-Net-101, Res-Net-152, InceptionV3 and Xception transfer learning networks. These deep learning networks were applied on CT images to assess the response to NAC. Performance was evaluated based on balanced_accuracy, accuracy, sensitivity and specificity classification metrics. A ViT transformer was applied to utilize the attention mechanism in order to increase the weight of important part image which leads to better discrimination between classes.ResultsAmongst the 117 LABC patients studied, 82 (70%) had clinical-pathological response and 35 (30%) had no response to NAC. The ViT transformer obtained the best performance range (accuracy = 71 ± 3% to accuracy = 77 ± 4%, specificity = 86 ± 6% to specificity = 76 ± 3%, sensitivity = 56 ± 4% to sensitivity = 52 ± 4%, and balanced_accuracy=69 ± 3% to balanced_accuracy=69 ± 3%) depending on the split ratio of train-data and test-data. Xception network obtained the second best results (accuracy = 72 ± 4% to accuracy = 65 ± 4, specificity = 81 ± 6% to specificity = 73 ± 3%, sensitivity = 55 ± 4% to sensitivity = 52 ± 5%, and balanced_accuracy = 66 ± 5% to balanced_accuracy = 60 ± 4%). The worst results were obtained using VGG-16 transfer learning network.ConclusionDeep learning networks in conjunction with CT imaging are able to predict the tumor response to NAC for patients with LABC prior to start. A ViT transformer could obtain the best performance, which demonstrated the importance of attention mechanism.

Toward explainable AI in radiology: Ensemble-CAM for effective thoracic disease localization in chest X-ray images using weak supervised learning

Chest X-ray (CXR) imaging is widely employed by radiologists to diagnose thoracic diseases. Recently, many deep learning techniques have been proposed as computer-aided diagnostic (CAD) tools to assist radiologists in minimizing the risk of incorrect diagnosis. From an application perspective, these models have exhibited two major challenges: (1) They require large volumes of annotated data at the training stage and (2) They lack explainable factors to justify their outcomes at the prediction stage. In the present study, we developed a class activation mapping (CAM)-based ensemble model, called Ensemble-CAM, to address both of these challenges via weakly supervised learning by employing explainable AI (XAI) functions. Ensemble-CAM utilizes class labels to predict the location of disease in association with interpretable features. The proposed work leverages ensemble and transfer learning with class activation functions to achieve three objectives: (1) minimizing the dependency on strongly annotated data when locating thoracic diseases, (2) enhancing confidence in predicted outcomes by visualizing their interpretable features, and (3) optimizing cumulative performance via fusion functions. Ensemble-CAM was trained on three CXR image datasets and evaluated through qualitative and quantitative measures via heatmaps and Jaccard indices. The results reflect the enhanced performance and reliability in comparison to existing standalone and ensembled models.

EfficientNet Transfer Learning Approach for Multi-Class Brain Tumor Classification

- It can be difficult for radiologists to accurately diagnose brain tumors in the medical field. When radiologists manually identify tumors, it can result in incorrect treatment planning and medical errors. The Miami Neuroscience Center states that there are 120 different kinds of brain tumors that can impact an individual's brain. Brain tumors that are most common and dangerous include glioma, meningioma, and pituitary tumors. In this work, we developed an online diagnostic tool to accurately classify brain tumors, including Astrocytoma, Glioma, Meningioma, Neurocytoma and Pituitary. Additionally, our program is capable of classifying a normal brain. We added CT and MRI scan images to improve cross-modality and model capability. We employed the CNN based neural network architecture known as EfficientNet. With this model, we achieved over 93.4% accuracy. Our application will help the radiologist identify tumors quickly and promote better treatment planning. Key words: Deep Learning, Transfer Learning, CNN, MRI, CT, Multi-Class Classification.

Enabling near-real-time safety glove detection through edge computing and transfer learning: comparative analysis of edge and cloud computing-based methods

PurposeComputer vision and deep learning (DL) methods have been investigated for personal protective equipment (PPE) monitoring and detection for construction workers’ safety. However, it is still challenging to implement automated safety monitoring methods in near real time or in a time-efficient manner in real construction practices. Therefore, this study developed a novel solution to enhance the time efficiency to achieve near-real-time safety glove detection and meanwhile preserve data privacy.Design/methodology/approachThe developed method comprises two primary components: (1) transfer learning methods to detect safety gloves and (2) edge computing to improve time efficiency and data privacy. To compare the developed edge computing-based method with the currently widely used cloud computing-based methods, a comprehensive comparative analysis was conducted from both the implementation and theory perspectives, providing insights into the developed approach’s performance.FindingsThree DL models achieved mean average precision (mAP) scores ranging from 74.92% to 84.31% for safety glove detection. The other two methods by combining object detection and classification achieved mAP as 89.91% for hand detection and 100% for glove classification. From both implementation and theory perspectives, the edge computing-based method detected gloves faster than the cloud computing-based method. The edge computing-based method achieved a detection latency of 36%–68% shorter than the cloud computing-based method in the implementation perspective. The findings highlight edge computing’s potential for near-real-time detection with improved data privacy.Originality/valueThis study implemented and evaluated DL-based safety monitoring methods on different computing infrastructures to investigate their time efficiency. This study contributes to existing knowledge by demonstrating how edge computing can be used with DL models (without sacrificing their performance) to improve PPE-glove monitoring in a time-efficient manner as well as maintain data privacy.