Transfer Learning Mechanism Research Articles

Modified LeNet-5 Architecture to Classify High Variety of Tourism Object: A Case Study of Tourism Object for Education in Tinalah Village

This research aims to modify a CNN (Convolutional Neural Network) based on LeNet-5 to reduce overfitting in a Tinalah Tourism Village dataset object detection. Tinalah Tourism Village has many objects that can be identified for tourism education and enhanced tourist experience. While these objects, spread across the different sites of Tinalah do vary, some share similarities in their histogram patterns. Visually, if the size of a picture is reduced in the LeNet-5 ‘preferred size’ feature, it will inevitably lose some of its information, making pictures too similar reducing accuracy. In order to learn and classify objects, this research performs a modification on LeNet-5 architecture to provide a better performance geared toward larger input imaging. The previous state-of-the-art architecture showed an overfitting performance where the training accuracy performed too much better than the testing accuracy in our dataset. We brought in a dropout layer to reduce overfitting, increase the dense layer's size, and add a convolution layer. We then compared the modified LeNet-5 with other state-of-the art architecture, such as LeNet-5 and AlexNet. Results showed that a modified LeNet-5 outperformed other architectures, especially in performing accuracy for testing the Tinalah dataset, reaching 0.913 or (91,3 %). This research discusses the dataset, the modified LeNet-5 architecture, and performance comparison between state-of-the-art CNN architecture. Our CNN architecture can be developed by involving a transfer learning mechanism to provide greater accuracy for further research.

Research on Spider Recognition Technology Based on Transfer Learning and Attention Mechanism

Methods such as transfer learning and attention mechanisms play an important role in small-sample image classification tasks. However, the conventional transfer method retains too much prior knowledge of the source domain and cannot learn the feature information of the target domain well. At the same time, it is difficult for the neural network model to find discriminative features and locate key feature regions, and it is easily interfered with by information such as complex backgrounds. Spiders usually appear symmetrical, but they are not perfectly symmetrical. How to accurately classify spider images depends on how to make the model focus on the key features for recognizing spiders in these symmetrical and asymmetrical regions. In view of the above problems, in this paper, we propose ECSM-ResNet-50, a model for small-sample spider image classification. The model fuses channel and spatial information and pays attention to the correlation between different locations in the input data. The Efficient Channel Attention (ECA) mechanism and the spatial attention mechanism were added to the model, and the self-attention mechanism was added to the end of the model. ECSM-ResNet-50 was constructed and trained on a small-sample spider data set (SPIDER9-IMAGE) using a layer-by-layer fine-tuning transfer learning strategy. Compared with ResNet-50, ECSM-ResNet-50 improved the average accuracy of nine species of spider recognition by 1.57% to 90.25%. This study contributes to the field of small-sample image recognition.

Computer Aided Diagnosis for COVID-19 in CT Images Utilizing Transfer Learning and Attention Mechanism

Computer Aided Diagnosis for COVID-19 in CT Images Utilizing Transfer Learning and Attention Mechanism

A NARX network optimized with an adaptive weighted square-root cubature Kalman filter for the dynamic state of charge estimation of lithium-ion batteries

Due to the high nonlinearities and unstable working conditions, accurately estimating the state of charge (SOC) by the battery management system (BMS) is a major challenge in ensuring the safety and reliability of lithium-ion batteries in electric vehicles. This paper presents a deep learning network, a nonlinear autoregressive model with exogenous inputs (NARX) network with a closed-loop architecture and transfer learning mechanism, which is optimized using a proposed adaptive weighted square-root cubature Kalman filter (AWSCKF) with a moving sliding window and an adaptive weighing coefficient for SOC estimation of lithium-ion batteries. The proposed AWSCKF method is established through square-root and cubature updates to optimize the statistical value of the state estimate, error covariance, and measurement noise covariance matrices, with the ability to incorporate high nonlinearities to filter out the noise, stabilize, and optimize the final SOC. To evaluate the effectiveness of the optimized NARX network and verify the proposed AWSCKF method, battery tests are carried out using a lithium cobalt oxide battery at various charge-discharge rates and a lithium nickel cobalt manganese oxide battery at temperatures of 0 and 45 °C under five complex working conditions. The SOC accuracy of lithium-ion batteries is enhanced by the hybrid method estimation process, which is based on sensitivity analysis and adaptation to various working conditions. The comprehensive results show that the proposed NARX-AWSCKF model achieves the overall best mean absolute error, root mean square error, and mean absolute percentage error values of 0.07293%, 0.0912%, and 0.40356%, respectively, under various complex conditions. By effectively utilizing battery domain knowledge for real-world BMS applications, the proposed model outperforms other existing methods in terms of high effectiveness, robustness, and potential to boost the NARX performance.

Transfer learning-based adaptive recursive neural network for short-term non-stationary building heating load prediction

Building energy consumption is a non-stationary time series, and its distribution law changes over time. Traditional machine-learning models are prone to model shift, which leads to a reduction of their prediction accuracy when applied to building energy consumption forecasting. Therefore, in this paper, an adaptive neural network framework is proposed for non-stationary building energy consumption prediction based on transfer learning, in which the training datasets are divided into the most dissimilar periods, and based on the transfer learning mechanism, the different periods of data are learned to obtain the minimum overall loss to improve the model generalization. Taking LSTM and GRU models as examples, adaptive long short-term memory (adaLSTM) and adaptive gated recurrent unit (adaGRU) building energy consumption prediction models are established. The models were trained and verified using heating load data from Xi'an, China. The results show that compared with LSTM model, the coefficient of determination, root mean square error, the coefficient of variation of the root mean squared error and mean absolute error of the adaLSTM model were improved by 0.61%, 37.78%, 38.05% and 30.69%, respectively, and the over-fitting degree was reduced by 227.7%. Compared with the traditional GRU model, the corresponding evaluation indexes of adaGRU were improved by 2.50%, 70.58%, 70.64% and 68.83%, respectively, and the over-fitting degree was improved by 505.7% points. The adaptive recurrent neural network framework proposed in this paper is a generalized approach which can be applied to other non-stationary time series prediction models.

Facial expression recognition for a seven-class small and medium-sized dataset based on transfer learning CNNs

As one of the most common biological information for human to express their emotions, facial expression plays an important role in biological research, psychological analysis and even human-computer interaction in the computer field. It is very important to use the efficient computing and processing power of computers to realize automatic facial expression recognition. However, in the research process of this field, most people's technology and model achievements are based on large datasets, which lack the universality of small and medium-sized datasets. Therefore, this project provides an optimization model on a specific seven-class small and medium face image dataset and provides a possible technical optimization reference direction for facial expression recognition models on similar small and medium-sized datasets. During the experiment, the training performance of VGG16 and MobileNet is compared. A comparative experiment is set up to observe the effect of transfer learning mechanism on training results. The results show that transfer learning has a significant effect on the performance of the model, and the accuracy of the optimal test set is more than 90%. Regardless of whether transfer learning mechanism is used, the training performance of VGG16 model structure is better than that of MobileNet structure in the same dataset.

A hybrid FCN-BiGRU with transfer learning for low-velocity impact identification on aircraft structure

In the work process of an aircraft, its external structure frequently encounters some low-velocity impact events, resulting in some barely visible impact damages inside the structure. Therefore, impact localization and reconstruction are critical for the structure’s health monitoring and reliability analysis. However, traditional inversion methods usually perform poorly in identifying random impact loads due to a series of problems. Inspired by the superior model properties of deep learning algorithms, a novel feature learning-based technique for impact load reconstruction and localization is presented. The proposed method consists of two parts, the first of which is utilized to localize the impact loads and is referred to as one-dimensional full convolution network. The other part aims at reconstructing the impact load and is called attention mechanism-full convolutional network-bidirectional gating recurrent unit-multilayer perceptron. Moreover, a transfer learning technique is introduced to optimize the model structure and process in light of the suggested innovative network model. The experimental piece of this research was an aircraft cutting segment whose structural surface was equipped with a number of piezoelectric sensors. These sensors received impact response signals with differences in amplitude from hammer strikes at various positions. In the work, the process of locating first and then reconstructing the impact load history was followed. And the influence of implementing the transfer learning mechanism on the performance of the impact load history reconstruction model was investigated as well. Three experiments and repeated cross-validation verified the efficacy of the proposed method. The findings indicated that the proposed technique could accurately and quickly detect impact loads at various locations. Simultaneously, it also allowed a perfect reconstruction of the impact loads of different impact energies.

Solving multi-task manufacturing cloud service allocation problems via bee colony optimizer with transfer learning

Industrial internet platform is regarded as an emerging infrastructure to increase the manufacturing efficiency via sharing resources located in multiple sites. Manufacturing cloud service allocation (MCSA) aims to assign available services to the interconnected subtasks of a complicated task such that some performance indices are optimized. Current studies on MCSA are single-task-oriented and fail to exploit the shared task-solving experiences to jointly optimize a group of tasks with enhanced solution quality and search speed. This work considers the joint optimization of multiple MCSA problems in a parallel fashion via cross-task transfer learning mechanism, and two novel transfer learning strategies are embedded into the framework of bee colony algorithm to make the best use of cross-task helpful knowledge when resolving multi-task MCSA. The first one is to design an individual-dependent transfer learning mechanism to govern the probability of whether a bee to perform intra-task self-evolution or cross-task knowledge transfer, which adaptively regulates the search behavior of each bee according to its state. The second one is to select the potential bees from foreign tasks for knowledge exchange with the aid of anomaly detection mechanism. The proposed optimizer is extensively examined on different scales of MCSA instances in multi-task scenario. Experimental results confirm the performance advantage of our proposal in comparison with other state-of-the-art peers.

Differential evolution-based transfer rough clustering algorithm

Due to well processing the uncertainty in data, rough clustering methods have been successfully applied in many fields. However, when the capacity of the available data is limited or the data are disturbed by noise, the rough clustering algorithms always cannot effectively explore the structure of the data. Furthermore, rough clustering algorithms are usually sensitive to the initialized cluster centers and easy to fall into local optimum. To resolve the problems mentioned above, a novel differential evolution-based transfer rough clustering (DE-TRC) algorithm is proposed in this paper. First, transfer learning mechanism is introduced into rough clustering and a transfer rough clustering framework is designed, which utilizes the knowledge from the related domain to assist the clustering task. Then, the objective function of the transfer rough clustering algorithm is optimized by using the differential evolution algorithm to enhance the robustness of the algorithm. It can overcome the sensitivity to initialized cluster centers and meanwhile achieve the global optimal clustering. The proposed algorithm is validated on different synthetic and real-world datasets. Experimental results demonstrate the effectiveness of the proposed algorithm in comparison with both traditional rough clustering algorithms and other state-of-the-art clustering algorithms.

A Lightweight Transfer Learning Model with Pruned and Distilled YOLOv5s to Identify Arc Magnet Surface Defects

Surface defects in arc magnets constitute the main culprit for performance degradation and safety hazards in permanent magnet motors. Machine-vision methods offer the possibility to identify surface defects automatically. However, the current methods still do not adequately solve the problems of low identification accuracy, excessive dependency on training data, and sizeable computational complexity. This paper proposes a lightweight YOLOv5s-based transfer learning model with network pruning and knowledge distillation to address these issues. Our model was derived from a pre-trained YOLOv5s for general object detection. A transfer learning mechanism was designed to obtain the optimal surface defect identification accuracy of the model from fewer training samples. Network pruning and knowledge distillation were combined to compress the transferred model. The transferred model serves as the teacher model of knowledge distillation, while its pruned model acts as the student model. To weaken the loss of the accuracy after model compression, a new λ factor was introduced into the confidence loss function of the student model to increase the sensitivity of identifying the defects. The experimental results show that our model’s performance is higher than other regular lightweight models. The identification accuracy for different defective arc magnets could reach 100%, the model size could achieve 1.921 MB, and the average inference time was 9.46 ms. Our model also has high accuracy in other defect identification applications besides arc magnets.

Region Segmentation of Whole-Slide Images for Analyzing Histological Differentiation of Prostate Adenocarcinoma Using Ensemble EfficientNetB2 U-Net with Transfer Learning Mechanism.

Recent advances in computer-aided detection via deep learning (DL) now allow for prostate cancer to be detected automatically and recognized with extremely high accuracy, much like other medical diagnoses and prognoses. However, researchers are still limited by the Gleason scoring system. The histopathological analysis involved in assigning the appropriate score is a rigorous, time-consuming manual process that is constrained by the quality of the material and the pathologist's level of expertise. In this research, we implemented a DL model using transfer learning on a set of histopathological images to segment cancerous and noncancerous areas in whole-slide images (WSIs). In this approach, the proposed Ensemble U-net model was applied for the segmentation of stroma, cancerous, and benign areas. The WSI dataset of prostate cancer was collected from the Kaggle repository, which is publicly available online. A total of 1000 WSIs were used for region segmentation. From this, 8100 patch images were used for training, and 900 for testing. The proposed model demonstrated an average dice coefficient (DC), intersection over union (IoU), and Hausdorff distance of 0.891, 0.811, and 15.9, respectively, on the test set, with corresponding masks of patch images. The manipulation of the proposed segmentation model improves the ability of the pathologist to predict disease outcomes, thus enhancing treatment efficacy by isolating the cancerous regions in WSIs.

Deep transfer learning for IDC breast cancer detection using fast AI technique and Sqeezenet architecture.

One of the most effective approaches for identifying breast cancer is histology, which is the meticulous inspection of tissues under a microscope. The kind of cancer cells, or whether they are cancerous (malignant) or non-cancerous, is typically determined by the type of tissue that is analyzed by the test performed by the technician (benign). The goal of this study was to automate IDC classification within breast cancer histology samples using a transfer learning technique. To improve our outcomes, we combined a Gradient Color Activation Mapping (Grad CAM) and image coloring mechanism with a discriminative fine-tuning methodology employing a one-cycle strategy using FastAI techniques. There have been lots of research studies related to deep transfer learning which use the same mechanism, but this report uses a transfer learning mechanism based on lightweight Squeeze Net architecture, a variant of CNN (Convolution neural network). This strategy demonstrates that fine-tuning on Squeeze Net makes it possible to achieve satisfactory results when transitioning generic features from natural images to medical images.

Adaptive Fusion and Transfer Learning for Enhanced E –Commerce Recommendations

Adaptive Fusion and Transfer Learning for Enhanced E-commerce Recommendations (AFETLER) is a holistic combination of some state-of-the-art recommendation frameworks. Most of the existing frameworks, be it Matrix Factorisation model, Deep learning-based model, Factorisation Machines and Two tower recommendation models, used in real world recommendation engines lack diversity and serendipity. Also, cold start problem is one that downgrade the recommendation algorithm. Defining the optimisation goal and closed feedback improvement loop is also something that is not given primary weightage.To overcome the above stated challenges and provide a novel approach to overcome those, we propose the AFETLER model. AFETLER incorporates an adaptive fusion technique that dynamically combines recommendations from several sources such as new arrivals forecast, Click-Through Rate (CTR) prediction, and popularity estimation. This fusion technique improves accuracy by giving varying weights to each source based on the user's context, history preferences, and current behaviour, resulting in highly relevant and tailored buying recommendations. To address the cold-start problem and improve recommendations for new items, AFETLER employs a transfer learning mechanism inspired by the Adversarial Two-tower Neural Network (ATNN) framework, utilizing knowledge from existing items' interactions to extract valuable features from new items' profiles. Furthermore, AFETLER leverages hierarchical attention methods to effectively capture user-item interaction patterns, boosting feature extraction and overall recommendation quality through dynamic weighing of interaction relevance.Proposed AFETLER model outperforms some of the existing recommendation models by an average 2%-5% on metrics such a precision, ROC-AUC, and Top-K recommendation accuracy. Further, we also see point reduction in Mean absolute error and mean squared error.

ICKSC :An Efficient Methodology for Predicting Kidney Stone From CT Kidney Image Dataset using Conventional Neural Networks

Chronic Kidney Diseases (CKD) has become one among the world wide health crisis and needs the associated efforts to prevent the complete organ damage. A considerable research effort has been put forward onto the effective separation and classification of kidney Stones from the kidney CT Images. Emerging machine learning along with deep learning algorithms have waved the novel paths of kidney stone detections. But these methods are proved to be laborious and its success rate is purely depends on the previous experiences. To achieve the better classification of kidney stone, this paper proposes a novel Intelligent CNN based Kidney Stone Classification (ICKSC) system which is based on transfer learning mechanism and incorporates 8 Layered CNN, densenet169_model, mobilenetv2_model, vgg19_model and xception_model. The extensive experimentation has been conducted to evaluate the efficacy of the recommended structure and matched with the other prevailing hybrid deep learning model. Experimentation demonstrates that the suggested model has showed the superior predominance over the other models and exhibited better performance in terms of training loss, accuracy, recall and precision.

An Automatic Premature Ventricular Contraction Recognition System Based on Imbalanced Dataset and Pre-Trained Residual Network Using Transfer Learning on ECG Signal

The development of automatic monitoring and diagnosis systems for cardiac patients over the internet has been facilitated by recent advancements in wearable sensor devices from electrocardiographs (ECGs), which need the use of patient-specific approaches. Premature ventricular contraction (PVC) is a common chronic cardiovascular disease that can cause conditions that are potentially fatal. Therefore, for the diagnosis of likely heart failure, precise PVC detection from ECGs is crucial. In the clinical settings, cardiologists typically employ long-term ECGs as a tool to identify PVCs, where a cardiologist must put in a lot of time and effort to appropriately assess the long-term ECGs which is time consuming and cumbersome. By addressing these issues, we have investigated a deep learning method with a pre-trained deep residual network, ResNet-18, to identify PVCs automatically using transfer learning mechanism. Herein, features are extracted by the inner layers of the network automatically compared to hand-crafted feature extraction methods. Transfer learning mechanism handles the difficulties of required large volume of training data for a deep model. The pre-trained model is evaluated on the Massachusetts Institute of Technology-Beth Israel Hospital (MIT-BIH) Arrhythmia and Institute of Cardiological Technics (INCART) datasets. First, we used the Pan-Tompkins algorithm to segment 44,103 normal and 6423 PVC beats, as well as 106,239 normal and 9987 PVC beats from the MIT-BIH Arrhythmia and IN-CART datasets, respectively. The pre-trained model employed the segmented beats as input after being converted into 2D (two-dimensional) images. The method is optimized with the using of weighted random samples, on-the-fly augmentation, Adam optimizer, and call back feature. The results from the proposed method demonstrate the satisfactory findings without the using of any complex pre-processing and feature extraction technique as well as design complexity of model. Using LOSOCV (leave one subject out cross-validation), the received accuracies on MIT-BIH and INCART are 99.93% and 99.77%, respectively, suppressing the state-of-the-art methods for PVC recognition on unseen data. This demonstrates the efficacy and generalizability of the proposed method on the imbalanced datasets. Due to the absence of device-specific (patient-specific) information at the evaluating stage on the target datasets in this study, the method might be used as a general approach to handle the situations in which ECG signals are obtained from different patients utilizing a variety of smart sensor devices.

Automatic video censoring system using deep learning

<span lang="EN-US">Due to the extensive use of video-sharing platforms and services, the amount of such all kinds of content on the web has become massive. This abundance of information is a problem controlling the kind of content that may be present in such a video. More than telling if the content is suitable for children and sensitive people or not, figuring it out is also important what parts of it contains such content, for preserving parts that would be discarded in a simple broad analysis. To tackle this problem, a comparison was done for popular image deep learning models: MobileNetV2, Xception model, InceptionV3, VGG16, VGG19, ResNet101 and ResNet50 to seek the one that is most suitable for the required application. Also, a system is developed that would automatically censor inappropriate content such as violent scenes with the help of deep learning. The system uses a transfer learning mechanism using the VGG16 model. The experiments suggested that the model showed excellent performance for the automatic censoring application that could also be used in other similar applications.</span>

Multi-donor Neural Transfer Learning for Genetic Programming

Genetic programming (GP), for the synthesis of brand new programs, continues to demonstrate increasingly capable results towards increasingly complex problems. A key challenge in GP is how to learn from the past so that the successful synthesis of simple programs can feed into more challenging unsolved problems. Transfer Learning (TL) in the literature has yet to demonstrate an automated mechanism to identify existing donor programs with high-utility genetic material for new problems, instead relying on human guidance. In this article we present a transfer learning mechanism for GP which fills this gap: we use a Turing-complete language for synthesis, and demonstrate how a neural network (NN) can be used to guide automated code fragment extraction from previously solved problems for injection into future problems. Using a framework which synthesises code from just 10 input-output examples, we first study NN ability to recognise the presence of code fragments in a larger program, then present an end-to-end system which takes only input-output examples and generates code fragments as it solves easier problems, then deploys selected high-utility fragments to solve harder ones. The use of NN-guided genetic material selection shows significant performance increases, on average doubling the percentage of programs that can be successfully synthesised when tested on two different problem corpora, compared with a non-transfer-learning GP baseline.

Multi-surface defect detection for universal joint bearings via multimodal feature and deep transfer learning

ABSTRACT Surface defect detection by machine vision has received increased attention concerning the quality control of universal joint bearings (UJB). The defect distribution and counting information are important for product quality optimisation. However, vision defect detection for UJB remains a challenging task due to the diversity of background textures and defect characters on multiple surfaces. In this study, a multi-surface defect detection (MSDD) method consisting of region segmentation, feature extraction and detection is proposed. First, defect regions are accurately localised by the proposed adaptive defect region segmentation algorithm, which suppresses the inference of background variety. Then, a novel defect feature named multimodal fusion shape descriptor that integrates the global information and local information of defects is constructed to generate the discriminative defect representation. Finally, a defect feature extraction ability transfer strategy based on the transfer learning mechanism is proposed to address the problem of insufficient defect samples. The experimental results show that our method achieves the best accuracy of 94.8% macro-F1 and processes 28 defect images per second. Besides, the application effect in the practical production line indicates that our method meets the accuracy and real-time requirements of MSDD for UJB.

NEURAL NETWORK BASED REAL TIME PNEUMONIA DETECTION USING TRANSFER LEARNING AND IMAGE AUGMENTATION

This paper aims to identify the detection of pneumonia disease using chest x-ray images by applying deep learning methods. Deep learning methods have tracked down their applications in different areas, and they are in effect broadly utilized in clinical medicines and diagnostics. To analyze viral/bacterial infections such as pneumonia, the assessment of chest X-ray images is frequently used, and the productivity of diagnosis can be altogether improved with the utilization of Computer-Aided Diagnostic (CAD) frameworks. Deep learning method such as Convolutional Neural Network (CNN) architecture is utilized in this paper for the characterization of chest X-ray images to analyze pneumonia. We have used the chest X-ray image dataset from Kaggle consisting of 4110 images. Image augmentations were performed on the dataset to oversample the dataset for the model to perform better. Then, at that point, we have built a custom CNN model and, also, we have utilized the transfer learning mechanism with CNN by using MobileNetV2 as the base model for the image classification problems. The average classification accuracy for our proposed CNN and MobileNetV2 based transfer learning method was 97%, and 97% for unbalanced and 97%, and 97%, for balanced datasets respectively. The satisfactory outcome of both models can significantly improve the accuracy and speed of pneumonia diagnosis. This would be very helpful in this pandemic situation in developing countries with limited resources and capabilities in the healthcare sector.

An optimized long short-term memory-weighted fading extended Kalman filtering model with wide temperature adaptation for the state of charge estimation of lithium-ion batteries

Accurate state of charge (SOC) estimation at different operating temperatures is essential for the reliable and safe operation of battery management systems (BMS) for lithium-ion batteries in electric vehicles (EVs). In this paper, an optimized long-short-term memory-weighted fading extended Kalman filtering (LSTM-WFEKF) model with wide temperature adaptation is proposed as a temperature-conditioned model for SOC estimation. Firstly, the input datasets are categorized based on the operating temperatures for EVs in the United States Advanced Battery Consortium manual: cold (−10 °C), normal (25 °C), and hot (50 °C) temperatures and optimized with an attention mechanism for faster training of the LSTM model to cross-train and test to specifically study the effects of temperature on the SOC estimation through a transfer learning mechanism. Secondly, the SOC estimated by the LSTM model is input into a WFEKF method, which introduces adaptive weighing and fading factors to correct, denoise, and optimize the final SOC for each temperature variation under complex working conditions. Finally, the results show that the training and testing temperatures have distinctive SOC effects using the LSTM model. Also, the proposed LSTM-WFEKF model estimates the SOC with overall best mean absolute error (MAE), root mean square error (RMSE), and R-squared (R2) values of 0.0697%, 0.0784%, and 99.9965%, respectively, under different temperatures and complex working conditions, which is optimal compared to other existing models. Based on the MAE, RMSE, and R2 values under different operating temperatures and complex working conditions, this paper concludes that the 25 °C training dataset ensures a more accurate SOC estimation. Meanwhile, the −10 °C and 50 °C training datasets cause more and less noisy estimates, respectively. The proposed LSTM-WFEKF model has wide temperature and working condition adaptability for real-time BMS applications in EVs.