Deep Transfer Learning Research Articles

Fault diagnosis of bearings under variable operating conditions based on domain adaptation

Aiming at the problem of low fault recognition rate caused by different distribution of training samples and test samples and imbalance of various fault data in bearing fault diagnosis, a domain adaptive fault diagnosis method based on improved residual network ( ResNet ) is designed. In the first layer of the diagnosis network, a multi-dimensional convolution structure is used for feature extraction to obtain fault feature information of different dimensions; the local maximum mean difference (LMMD) is used in the domain adaptive layer to align the distribution of the source domain and the target domain to obtain more fine-grained information; the class balance loss function ( CBLoss ) is used to solve the training problem of unbalanced data, and the Adam optimization network is used to realize fault diagnosis. Experimental results show that the proposed improved method can achieve higher diagnosis results under the imbalance of fault sample categories. Experimental verification is carried out on two bearing data sets and collected wind turbine data. The results show that the proposed improved method has certain advantages. The diagnostic performance of the proposed improved method in the case of unbalanced data samples is better than other deep transfer learning methods, and it can be used as an effective cross-condition fault analysis method.

Transfer learning-assisted modulation classification and OSNR estimation using constellation diagrams

Abstract A cost-effective deep transfer learning (TL) learned with characteristics derived from the constellation diagram is proposed for executing modulation classification and optical signal-to-noise ratio (OSNR) estimation for the future generation of elastic optical networks. The information acquired with the ImageNet dataset is transferred using pre-trained TL versions including VGG19, ResNet152V2, InceptionV3, and ResNet50 to identify various modulation formats and their corresponding OSNR. Adam optimizer is employed to ascertain the appropriate settings for the hyperparameters. Numerical results demonstrate that the proposed VGG19 method achieves the highest level of accuracy (100%) in the recognition of various modulation forms. To fulfil the requirements of actual use, OSNR monitoring is also explored, with a total precision equal to 95.2%. Furthermore, a thorough investigation of the impact of training data dimensions, on TL performance is conducted. The outcomes of the proposed method demonstrate that the suggested TL-based algorithms are more accurate and need much less training and testing time than non-TL approaches.

Natural language processing‐based deep transfer learning model across diverse tabular datasets for bond strength prediction of composite bars in concrete

AbstractAs conventional machine learning models often struggle with scarcity and structural variation of training data, this paper proposes a novel regression transfer learning framework called transferable tabular regressor (TransTabRegressor) to address this challenge. The TransTabRegressor integrates natural language processing (NLP) for feature encoding, transformer for enhanced feature representation, and deep learning (DL) for robust modeling, facilitating effective transfer learning across tabular datasets using reducing input parameters. By leveraging the NLP data processor, the framework embeds both parameter names and values, enabling it to recognize and adapt to different expressions of similar parameters. For instance, the bond strength of fiber‐reinforced polymer (FRP) bars embedded in ultra‐high‐performance concrete (UHPC) is critical for ensuring the integrity of FRP‐UHPC structures. While pullout tests are widely adopted for their simplicity to generate substantial data, beam tests provide a closer approximation to actual stress conditions but are more complex thus resulting in limited data size. As a verification, the framework is applied to predict the bond strength of FRP bars embedded in UHPC using limited beam test data. A pre‐trained model is first established using 479 pieces of pullout test data. Subsequently, two transfer learning models are developed by fine‐tuning on 115 pieces of beam test data, where 66 correspond to concrete splitting failure and 49 correspond to pullout failure. For comparative analysis, XGBoost and neural network models are directly trained on the beam test data. Evaluation results demonstrate that the transfer learning models achieve significantly improved prediction accuracy and generalization capability. This study significantly highlights the effectiveness of the proposed TransTabRegressor in handling data scarcity and variability in input parameters across various engineering applications.

A novel ensemble approach with deep transfer learning for accurate identification of foodborne bacteria from hyperspectral microscopy

The detection of foodborne bacteria is critical in ensuring both consumer safety and food safety. If these pathogens are not properly identified, it can lead to dangerous cross-contamination. One of the most common methods for classifying bacteria is through the examination of Hyperspectral microscope imaging (HMI). A widely used technique for measuring microbial growth is microscopic cell counting. HMI is a laborious and expensive process, producing voluminous data and needing specialized equipment, which might not be widely available. Machine learning (ML) methods are now frequently utilized to automatically interpret data from hyperspectral microscopy. The objective of our study is to devise a technique that employs deep transfer learning to address the challenge of limited data and utilizes four base classifiers - InceptionResNetV2, MobileNet, ResNet101V2, and Xception - to create an ensemble-based classification model for distinguishing live and dead bacterial cells of six pathogenic strains. In order to determine the optimal weights for the base classifiers, a Powell's optimization method was utilized in conjunction with a weighted average ensemble (WAVE) technique. We carried out an extensive experimental study to verify the efficiency of our proposed ensemble model on live and dead cell images of six different foodborne bacteria. In order to gain a better understanding of the regions, we performed a Grad-CAM analysis to explain the predictions made by our model. Through a series of experiments, our proposed framework has proven its capacity to effectively and precisely detect numerous bacterial pathogens. Specifically, it achieved a perfect identification rate of 100 % for Escherichia coli (EC), Listeria innocua (LI), and Salmonella Enteritidis (SE), while achieving rates of 96.30 % for Salmonella Typhimurium (ST), 87.13 % for Staphylococcus aureus (SA), and 94.12 % for Salmonella Heidelberg (SH). As a result, it can be considered as an effective tool for the identification of foodborne pathogens, due to its high level of efficiency.

An EfficientNet integrated ResNet deep network and explainable AI for breast lesion classification from ultrasound images

AbstractBreast cancer is one of the major causes of deaths in women. However, the early diagnosis is important for screening and control the mortality rate. Thus for the diagnosis of breast cancer at the early stage, a computer‐aided diagnosis system is highly required. Ultrasound is an important examination technique for breast cancer diagnosis due to its low cost. Recently, many learning‐based techniques have been introduced to classify breast cancer using breast ultrasound imaging dataset (BUSI) datasets; however, the manual handling is not an easy process and time consuming. The authors propose an EfficientNet‐integrated ResNet deep network and XAI‐based framework for accurately classifying breast cancer (malignant and benign). In the initial step, data augmentation is performed to increase the number of training samples. For this purpose, three‐pixel flip mathematical equations are introduced: horizontal, vertical, and 90°. Later, two pre‐trained deep learning models were employed, skipped some layers, and fine‐tuned. Both fine‐tuned models are later trained using a deep transfer learning process and extracted features from the deeper layer. Explainable artificial intelligence‐based analysed the performance of trained models. After that, a new feature selection technique is proposed based on the cuckoo search algorithm called cuckoo search controlled standard error mean. This technique selects the best features and fuses using a new parallel zero‐padding maximum correlated coefficient features. In the end, the selection algorithm is applied again to the fused feature vector and classified using machine learning algorithms. The experimental process of the proposed framework is conducted on a publicly available BUSI and obtained 98.4% and 98% accuracy in two different experiments. Comparing the proposed framework is also conducted with recent techniques and shows improved accuracy. In addition, the proposed framework was executed less than the original deep learning models.

Nested deep transfer learning for modeling of multilayer thin films

Nested deep transfer learning for modeling of multilayer thin films

Computer vision and deep transfer learning for automatic gauge reading detection

This manuscript proposes an automatic reading detection system for an analogue gauge using a combination of deep learning, machine learning, and image processing. The study suggests image-processing techniques in manual analogue gauge reading that include generating readings for the image to provide supervised data to address difficulties in unsupervised data in gauges and to achieve better accuracy using DenseNet 169 compared to other approaches. The model uses artificial intelligence to automate reading detection using deep transfer learning models like DenseNet 169, InceptionNet V3, and VGG19. The models were trained using 1011 labeled pictures, 9 classes, and readings from 0 to 8. The VGG19 model exhibits a high training precision of 97.00% but a comparatively lower testing precision of 75.00%, indicating the possibility of overfitting. On the other hand, InceptionNet V3 demonstrates consistent precision across both datasets, but DenseNet 169 surpasses other models in terms of precision and generalization capabilities.

Deep learning predicted perceived age is a reliable approach for analysis of facial ageing: A proof of principle study.

Perceived age (PA) has been associated with mortality, genetic variants linked to ageing and several age-related morbidities. However, estimating PA in large datasets is laborious and costly to generate, limiting its practical applicability. To determine if estimating PA using deep learning-based algorithms results in the same associations with morbidities and genetic variants as human-estimated perceived age. Self-supervised learning (SSL) and deep feature transfer (DFT) deep learning (DL) approaches were trained and tested on human-estimated PAs and their corresponding frontal face images of middle-aged to elderly Dutch participants (n = 2679) from a population-based study in the Netherlands. We compared the DL-estimated PAs with morbidities previously associated with human-estimated PA as well as genetic variants in the gene MC1R; we additionally tested the PA associations with MC1R in a new validation cohort (n = 1158). The DL approaches predicted PA in this population with a mean absolute error of 2.84 years (DFT) and 2.39 years (SSL). In the training-test dataset, we found the same significant (p < 0.05) associations for DL PA with osteoporosis, ARHL, cognition, COPD and cataracts and MC1R, as with human PA. We also found a similar but less significant association for SSL and DFT PAs (0.69 and 0.71 years per allele, p = 0.008 and 0.011, respectively) with MC1R variants in the validation dataset as that found with human, SSL and DFT PAs in the training-test dataset (0.79, 0.78 and 0.71 years per allele respectively; all p < 0.0001). Deep learning methods can automatically estimate PA from facial images with enough accuracy to replicate known links between human-estimated perceived age and several age-related morbidities. Furthermore, DL predicted perceived age associated with MC1R gene variants in a validation cohort. Hence, such DL PA techniques may be used instead of human estimations in perceived age studies thereby reducing time and costs.

Enhancing the security in IoT and IIoT networks: An intrusion detection scheme leveraging deep transfer learning

The Internet of Things (IoT) networks, which are defined by their interconnected devices and data streams are an expanding attack surface for cyber adversaries. Industrial Internet of Things (IIoT) is a subset of IoT and has significant importance in-terms of security. Robust intrusion detection systems (IDS) are essential for protecting these critical infrastructures. Our research suggests a novel approach to the detection of anomalies in IoT and IIoT networks that leverages the capabilities of deep transfer learning. Our methodology begins with the EdgeIIoT dataset, which serves as the basis for our data analysis. We convert the data into an appropriate image format to enable Convolutional Neural Network (CNN)-based processing. The hyper-parameters of individual machine learning models are subsequently optimized using a Random Search algorithm. This optimization phase optimizes the performance of each model by modifying the hyper-parameters that are unique to the learning algorithms. The performance of each model is meticulously assessed subsequent to hyper-parameter optimization. The top-performing models are subsequently, strategically selected and combined using the ensemble technique. The IDS scheme’s overall detection accuracy and generalizability are improved by the integration of strengths from multiple models. The proposed scheme demonstrates significant effectiveness in identifying a broad spectrum of attacks, encompassing a total of 14 distinct attack types. This comprehensive detection capability contributes to a more secure and resilient IoT ecosystem. Furthermore, application of quantization to our best models reduces resource utilization significantly without compromising accuracy.

Detection of COVID-19 from chest X-rays using deep transfer learning

Objectives: As COVID-19 continues to wreak havoc around the world, effective screening methods for safeguarding public health and bringing the pandemic under control are urgently required. Accordingly, this study proposes a robust pipeline for training and evaluating deep learning predictive models for the detection of COVID-19 from chest X-rays. Methods: The pipeline incorporates multiple techniques to combat overfitting and optimize the prediction results, including data augmentation, Bayesian optimization, the selection of appropriate performance metrics, and the use of early stopping during model training. The proposed pipeline is applied to three common deep learning models: ResNet50, NASNet-A-Mobile, and Xception. Results: The experimental results obtained using the COVID-XRay-5K v3 dataset with 2084 training and 3100 test examples show that the three models achieve AU-PRC (area under precision-recall curve) scores of 0.977, 0.963 and 0.900, respectively; and AU-ROC (area under receiver operating characteristic) scores of 0.994, 0.996 and 0.981. Moreover, at a 98.00% recall (sensitivity), the models achieve high specificities of 97.53%, 97.60% and 86.00%, respectively. Conclusion: Overall, the results suggest that all three models provide a promising approach for the rapid and reliable detection of COVID-19.

Image classification of fabric defects using ResNet50 deep transfer learning in FastAI

Image classification of fabric defects using ResNet50 deep transfer learning in FastAI

An innovative deep reinforcement learning-driven cutting parameters adaptive optimization method taking tool wear into account

Tool wear is critically important for the optimization of cutting parameters. However, the increasing nature of tool wear presents challenges to traditional meta-heuristic cutting parameter optimization methods. To address this issue, we propose an innovative deep reinforcement learning-driven cutting parameters adaptive optimization method taking tool wear into account. More specifically, we use the Markov Decision Process to simulate the optimization process of cutting parameters. Firstly, an innovative deep transfer learning algorithm is used for monitoring tool wear. With the progress of tool wear, the proximal policy optimization method of the transformer with multi-head attention mechanism interacts with the processing environment through a process of trial and error, and accumulates a wealth of experience in selecting cutting parameters through the reward function. The deep reinforcement learning model has quickly discern the best cutting parameters, relying on real-time tool wear value. The experimental results show that the proposed method outperforms other algorithms.

Arc bubble edge detection method based on deep transfer learning in underwater wet welding.

The stability of arc bubble is a crucial indicator of underwater wet welding process. However, limited research exists on detecting arc bubble edges in such environments, and traditional algorithms often produce blurry and discontinuous results. To address these challenges, we propose a novel arc bubble edge detection method based on deep transfer learning for processing underwater wet welding images. The proposed method integrates two training stages: pre-training and fine-tuning. In the pre-training stage, a large source domain dataset is used to train VGG16 as a feature extractor. In the fine-tuning stage, we introduce the Attention-Scale-Semantics (ASS) model, which consists of a Convolutional Block Attention Module (CBAM), a Scale Fusion Module (SCM) and a Semantic Fusion Module (SEM). The ASS model is further trained on a small target domain dataset specific to underwater wet welding to fine-tune the model parameters. The CBAM can adaptively weight the feature maps, focusing on more crucial features to better capture edge information. The SCM training method maximizes feature utilization and simplifies training by combining multi-scale features. Additionally, the skip structure of SEM effectively mitigates semantic loss in the high-level network, enhancing the accuracy of edge detection. On the BSDS500 dataset and a self-constructed underwater wet welding dataset, the ASS model was evaluated against conventional edge detection models-Richer Convolutional Features (RCF), Fully Convolutional Network (FCN), and UNet-as well as state-of-the-art models LDC and TEED. In terms of Mean Absolute Error (MAE), accuracy, and other evaluation metrics, the ASS model consistently outperforms these models, demonstrating edge detection capabilities that are both effective and stable in detecting arc bubble edges in underwater wet welding images.

Voice recognition by deep transfer learning and vision transformers to secure voice authentication

Speech recognition is crucial for ensuring the security of personal devices and financial transactions. Attaining high accuracy and robustness in voice authentication is challenging due to the presence of voice and environmental variability. Recent advancements in the field of deep learning, particularly in transfer learning and visual transformers, have the potential to enhance voice recognition systems. This study employs advanced deep transfer learning techniques, including Vision Transformers (ViT), VGG16, and a customized Convolutional Neural Network (CNN), to enhance the accuracy and security of speech authentication. The objective is to evaluate and contrast various solutions' voice recognition and authentication accuracy. The experiment included 3000 voice samples, with an equal distribution of 1500 samples from male participants and 1500 from female participants. The dataset was used to train Vision Transformers, VGG16 with transfer learning, and a custom CNN. The models were assessed based on their accuracy in identifying and authenticating voice samples. The VGG16 model achieved the highest level of accuracy in speech recognition, with a precision rate of 95%. The Vision Transformer and custom CNN exhibited satisfactory performance. However, VGG16 demonstrated higher accuracy. The most accurate voice authentication model studied is the VGG16 model based on transfer learning. This study suggests that the security and reliability of voice recognition systems can be enhanced through the use of deep learning techniques.

Statistical Approaches Used in Studies Evaluating the Reliability of Autonomous Vehicles Based on Disengagements and Reaction Times

The rapid development of autonomous vehicle (AV) technology highlights the critical importance of enhancing the reliability of these vehicles. Due to the need to test the reliability of AVs, since 2014, the California Department of Motor Vehicles has permitted autonomous vehicle manufacturers to establish an AV Testing program, enabling them to test automated systems on the transportation network. With this, studies on the reliability of AVs have increased rapidly. The most emphasized issues regarding the reliability of AVs have been disengagements, accidents, and reaction times. In this study, disengagements and reaction times are categorized and explained in detail according to the data type, company, period, and statistical method. The data used in the studies cover the years 2014-2020. When examining studies on the reliability of AVs, until 2018, inferences were generally made using real data and descriptive statistics, particularly with methods such as correlation analysis and calculation of disengagements per mile, which investigates the relationship between distance traveled and disengagements. However, since 2018, machine learning has gained importance in evaluating AV reliability. It has been observed that regression, classification, and decision trees were frequently used during this period. Techniques such as deep transfer learning, text mining, and natural language processing also stand out. Furthermore, Software Reliability Growth Models are used to measure software reliability, playing an essential role in evaluating, analyzing, and improving the performance and reliability of AVs. This study aims to reveal the development and diversity of the statistical methods used to determine AV reliability. Additionally, this study aims to guide and provide insights to researchers in the field about the statistical approaches they can utilize.

Robust Face Recognition System Based on Deep Facial Feature Extraction and Machine Learning

Facial Recognition is a common challenge in research for security applications. Wide-ranging applications of this technology include biometrics, security data, access to controlled locations, smart cards, and surveillance systems. A practical approach to identifying individuals depends on factors such as partial face occlusion, illumination, age, expression, makeup, and poses. These complex face recognition variables affect most face recognition systems. In this paper, we propose a deep transfer learning for facial features based on the pose and illumination variations of celebrities’ faces in a free environment for face recognition. The new approach that combines deep learning methods for feature extraction with machine learning classification methods. Also, studies the performance of the pre-trained model (VGGFace2 with ResNet-50 model weights) for feature extraction with a Multilayer Perceptron Classifier (MLP), Decision Tree, and Bootstrap Aggregation (Bagging) to perform classification. The convolution neural network has lately achieved excellent advancement in facial recognition. The outcome indicated that VGGFace2 with MLP obtained more precision and provided the best results (Accuracy, F-measure, Recall, and Precision). The results for all metric for the proposed model is 99%. The benchmark for this model is a dataset of 105 people’s faces.

Optimizing wearable single-channel EEG sleep staging in a heterogeneous sleep-disordered population using transfer learning.

While various wearable EEG devices have been developed, performance evaluation of the devices and their associated automated sleep stage classification models is mostly limited to healthy subjects. A major barrier for applying automated wearable EEG sleep staging in clinical populations is the need for large-scale data for model training. We therefore investigated transfer learning as strategy to overcome limited data availability and optimize automated single-channel EEG sleep staging in people with sleep disorders. We acquired 52 single-channel frontopolar headband EEG recordings from a heterogeneous sleep-disordered population with concurrent PSG. We compared three model training strategies: 'pre-training' (i.e., training on a larger dataset of 901 conventional PSGs), 'training-from-scratch' (i.e., training on wearable headband recordings), and 'fine-tuning' (i.e., training on conventional PSGs, followed by training on headband recordings). Performance was evaluated on all headband recordings using 10-fold cross-validation. Highest performance for 5-stage classification was achieved with fine-tuning (κ = .778), significantly higher than with pre-training (κ = .769) and with training-from-scratch (κ = .733). No significant differences or systematic biases were observed with clinically relevant sleep parameters derived from PSG. All sleep disorder categories showed comparable performance. This study emphasizes the importance of leveraging larger available datasets through deep transfer learning to optimize performance with limited data availability. Findings indicate strong similarity in data characteristics between conventional PSG and headband recordings. Altogether, results suggest the combination of the headband, classification model, and training methodology can be viable for sleep monitoring in the heterogeneous clinical population.

Transferable Deep Learning Models for Accurate Ankle Joint Moment Estimation during Gait Using Electromyography

The joint moment is a key measurement in locomotion analysis. Transferable prediction across different subjects is advantageous for calibration-free, practical clinical applications. However, even for similar gait motions, intersubject variance presents a significant challenge in maintaining reliable prediction performance. The optimal deep learning models for ankle moment prediction during dynamic gait motions remain underexplored for both intrasubject and intersubject usage. This study evaluates the feasibility of different deep-learning models for estimating ankle moments using sEMG data to find an optimal intrasubject model against the inverse dynamic approach. We verified and compared the performance of 1302 intrasubject models per subject on 597 steps from seven subjects using various architectures and feature sets. The best-performing intrasubject models were recurrent convolutional neural networks trained using signal energy features. They were then transferred to realize intersubject ankle moment estimation.

A vision transformer-based deep transfer learning nomogram for predicting lymph node metastasis in lung adenocarcinoma.

Lymph node metastasis (LNM) plays a crucial role in the management of lung cancer; however, the ability of chest computed tomography (CT) imaging to detect LNM status is limited. This study aimed to develop and validate a vision transformer-based deep transfer learning nomogram for predicting LNM in lung adenocarcinoma patients using preoperative unenhanced chest CT imaging. This study included 528 patients with lung adenocarcinoma who were randomly divided into training and validation cohorts at a 7:3 ratio. The pretrained vision transformer (ViT) was utilized to extract deep transfer learning (DTL) feature, and logistic regression was employed to construct a ViT-based DTL model. Subsequently, the model was compared with six classical convolutional neural network (CNN) models. Finally, the ViT-based DTL signature was combined with independent clinical predictors to construct a ViT-based deep transfer learning nomogram (DTLN). The ViT-based DTL model showed good performance, with an area under the curve (AUC) of 0.821 (95% CI, 0.775-0.867) in the training cohort and 0.825 (95% CI, 0.758-0.891) in the validation cohort. The ViT-based DTL model demonstrated comparable performance to classical CNN models in predicting LNM, and the ViT-based DTL signature was then used to construct ViT-based DTLN with independent clinical predictors such as tumor maximum diameter, location, and density. The DTLN achieved the best predictive performance, with AUCs of 0.865 (95% CI, 0.827-0.903) and 0.894 (95% CI, 0845-0942), respectively, surpassing both the clinical factor model and the ViT-based DTL model (p<0.001). This study developed a new DTL model based on ViT to predict LNM status in lung adenocarcinoma patients and revealed that the performance of the ViT-based DTL model was comparable to that of classical CNN models, confirming that ViT was viable for deep learning tasks involving medical images. The ViT-based DTLN performed exceptionally well and can assist clinicians and radiologists in making accurate judgments and formulating appropriate treatment plans.

MSlocPRED: deep transfer learning-based identification of multi-label mRNA subcellular localization.

Subcellular localization of messenger ribonucleic acid (mRNA) is a universal mechanism for precise and efficient control of the translation process. Although many computational methods have been constructed by researchers for predicting mRNA subcellular localization, very few of these computational methods have been designed to predict subcellular localization with multiple localization annotations, and their generalization performance could be improved. In this study, the prediction model MSlocPRED was constructed to identify multi-label mRNA subcellular localization. First, the preprocessed Dataset 1 and Dataset 2 are transformed into the form of images. The proposed MDNDO-SMDU resampling technique is then used to balance the number of samples in each category in the training dataset. Finally, deep transfer learning was used to construct the predictive model MSlocPRED to identify subcellular localization for 16 classes (Dataset 1) and 18 classes (Dataset 2). The results of comparative tests of different resampling techniques show that the resampling technique proposed in this study is more effective in preprocessing for subcellular localization. The prediction results of the datasets constructed by intercepting different NC end (Both the 5' and 3' untranslated regions that flank the protein-coding sequence and influence mRNA function without encoding proteins themselves.) lengths show that for Dataset 1 and Dataset 2, the prediction performance is best when the NC end is intercepted by 35 nucleotides, respectively. The results of both independent testing and five-fold cross-validation comparisons with established prediction tools show that MSlocPRED is significantly better than established tools for identifying multi-label mRNA subcellular localization. Additionally, to understand how the MSlocPRED model works during the prediction process, SHapley Additive exPlanations was used to explain it. The predictive model and associated datasets are available on the following github: https://github.com/ZBYnb1/MSlocPRED/tree/main.