Transfer Learning-based Approach Research Articles

Multi-class classification of brain tumor types from MR images using EfficientNets

Accurate classification of the type of brain tumor plays an important role in the early diagnosis of the tumor which can be the difference between life and death. Magnetic Resonance Imaging (MRI) is commonly used to capture high-contrast grayscale images of the brain and is a non-invasive method for brain tumor diagnosis. Deep Learning (DL) using Convolutional Neural Networks (CNN) has revolutionized Computer Aided Diagnostic (CAD) systems by producing remarkable results for various medical imaging analysis tasks including brain tumor detection. In this research, a transfer learning-based fine-tuning approach for the classification of brain tumors into three classes i.e. glioma, meningioma, and pituitary tumor using EfficientNets is carried out. Five variants of pre-trained models from the EfficientNets family i.e. EfficientNetB0 – EfficientNetB4 are fine-tuned on publically available CE-MRI “Figshare – Brain Tumor Dataset”. The proposed method of fine-tuning pre-trained EfficientNet is carried out by first loading ImageNet weights to the EfficientNet model followed by the addition of several top layers and a fully connected layer for the classification of brain tumor types. Several experiments are carried out to analyze the robustness of the proposed fine-tuned EfficientNets in comparison to other pre-trained models. Moreover, the effect of data augmentation on the model’s test accuracy is also studied. Finally, the Grad-CAM visualization of the attention maps of the best model is presented that successfully highlights the tumorous region of the brain cell. The proposed method of fine-tuning pre-trained EfficientNet showed significant performance using EfficientNetB2 as a backbone achieving overall test accuracy, precision, recall/sensitivity, and F1-score of 98.86%, 98.65%, 98.77%, and 98.71% respectively. The proposed fine-tuned EfficientNetB2 is lightweight, computationally inexpensive (during training and inference), and generalizes well. The source code is available at https://github.com/FatimaZulfiqar/multi-class-brain-tumor-classification.

Deep transfer learning-based hierarchical adaptive remaining useful life prediction of bearings considering the correlation of multistage degradation

The multi-stage degradation process of bearings significantly affects the predicted performance of rotating machinery and equipment in long-term operation. However, the inherent correlations between degradation stages, and domain-invariant degradation features of multistage that affect the remaining useful life (RUL) are mostly ignored, leading to RUL prognostics deterioration under cross-working conditions. Therefore, a novel deep transfer learning-based hierarchical adaptive RUL prediction approach is applied to overcome this problem. Firstly, a novel multistage degradation (MD) division method is proposed with a combination of maximum mean discrepancy and statistical process analysis to accurately obtain the varied health indicators (HIs) with MD without setting any threshold. Then, the obtained MD features are fed into the residual network to automatically recognize the HIs and MD in response to multi-variability trend changes. Finally, a novel hierarchical adaptive RUL model based on a deep adaptive-transformer is proposed to transfer the learned domain-invariant degradation features of multistage from one to various working conditions. Meanwhile, hierarchical adaptive tuning (HAT) reduces the RUL model loss to guarantee a more accurate prediction of each degradation stage. To validate the proposed approach, extensive experiments were conducted in seven cross-domain cases on the XJTU-SY dataset to predict RUL. Comparison results indicate that the proposed approach is more accurate than other conventional methods because it considers the RUL of bearing at each stage.

Tire Condition Monitoring Using Transfer Learning-Based Deep Neural Network Approach.

Monitoring tire condition plays a deterministic role in the overall safety and economy of an automobile. The tire condition monitoring system (TCMS) alerts the driver of the vehicle if the inflation pressure of a particular tire decreases below a specific value. Owing to the high costs involved in realizing this system, most vehicles do not feature this technology as a standard. With highly robust and accurate sensors making their way into an increasing number of applications, obtaining signals of varied types (especially vibration signals) is becoming easier and more modularized. In addition, feature-based machine learning techniques that enable accurate responses to varied input conditions have sought greater scientific attention. However, deep learning is gradually finding greater applications pertaining to condition monitoring. One approach of deep learning is presented in this paper, which instantaneously monitors the vehicle tire condition. For this purpose, vibration signals were obtained through the rotation of the tire under different inflation pressure conditions using a low-cost microelectromechanical system (MEMS) accelerometer.

Deep ensemble transfer learning-based approach for classifying hot-rolled steel strips surface defects

Over the last few years, advanced deep learning-based computer vision algorithms are revolutionizing the manufacturing field. Thus, several industry-related hard problems can be solved by training these algorithms, including flaw detection in various materials. Therefore, identifying steel surface defects is considered one of the most important tasks in the steel industry. In this paper, we propose a deep learning-based model to classify six of the most common steel strip surface defects using the NEU-CLS dataset. We investigate the effectiveness of two state-of-the-art CNN architectures (MobileNet-V2 and Xception) combined with the transfer learning approach. The proposed approach uses an ensemble of two pre-trained state-of-the-art Convolutional Neural Networks, which are MobileNet-V2 and Xception. To perform a comparative analysis of the proposed architectures, several evaluation metrics are adopted, including loss, accuracy, precision, recall, F1-score, and execution time. The experimental results show that the proposed deep ensemble learning approach provides higher performance achieving an accuracy of 99.72% compared to MobileNet-V2 (98.61%) and Xception (99.17%) while preserving fast execution time and small models’ size.

Real-time plume tracking using transfer learning approach

Deep learning has been used to track the real-time flammable plume of natural gas. However, a large volume of high-fidelity data is required to train the deep learning model for sufficient accuracy in congested industrial environments, which can be computationally prohibitive. This study proposes a transfer learning-based variable-fidelity approach for real-time plume tracking. A Gaussian dispersion model was applied to efficiently generate a large volume of low-fidelity data, which is then used to pre-train the deep learning model. A limited number of high-fidelity simulations were conducted by solving the Navier-Stokes equation to fine-tune the pre-trained model. A case study demonstrated our proposed approach could reduce high-fidelity computations by 72% while ensuring prediction accuracy with R2=0.96 for released plume area estimation in congested chemical facilities. Optimal number of frozen layers, learning rate and the number of high-fidelity simulations required were determined to ensure adequate efficiency for this approach. This study provides an efficient alternative to improve the generalization of deep learning for real-time plume area estimation for large-scale congested chemical plants.

Transfer Learning-Based Intelligent Fault Detection Approach for the Industrial Robotic System

With increasing customer demand, industry 4.0 gained a lot of interest, which is based on smart factories. In smart factories, robotic components are vulnerable to failure due to various industrial operations such as assembly, manufacturing, and product handling. Timely fault detection and diagnosis (FDD) is important to keep the industrial operation smooth. Previously, only the unloaded-based FDD algorithms were considered for the industrial robotic system. In the industrial environment, the robot is working under various working conditions such as speeds, loads, and motions. Hence, to reduce the domain discrepancy between the lab scale and the real working environment, we conducted experimentations under various working conditions. For that purpose, an extensive experimental setup is prepared to perform a series of various experiments mimicking the real environmental condition. In addition, in previous research work, various machine learning (ML) and deep learning (DL) approaches were proposed for robotic arm component fault detection. However, various issues are related to the DL and ML approaches. The ML models are problem-specific, and complex in computations. The DL model needs a huge amount of data. The DL model is composed of various layers that have not been thoroughly explored; as a result, the fault detection model lacks a comprehensive explanation. To overcome these issues, the transfer learning (TL) model is considered with the diverse experimental scenarios. The main contribution is to increase the generalization capabilities of the robotic PHM in the context of previously available research work. For that purpose, the VGG16 model is used because of its autonomous feature extractions for fault classification. The data are collected under a variety of different operating conditions such as loadings, speeds, and motion patterns. The 1D signal is converted to a 2D signal (scalogram) to perform the TL model. The proposed approach shows effective fault detection performance and has the capabilities of generalization under variable working conditions.

Transfer-Learning-Based Estimation of the Remaining Useful Life of Heterogeneous Bearing Types Using Low-Frequency Accelerometers.

Deep learning approaches are becoming increasingly important for the estimation of the Remaining Useful Life (RUL) of mechanical elements such as bearings. This paper proposes and evaluates a novel transfer learning-based approach for RUL estimations of different bearing types with small datasets and low sampling rates. The approach is based on an intermediate domain that abstracts features of the bearings based on their fault frequencies. The features are processed by convolutional layers. Finally, the RUL estimation is performed using a Long Short-Term Memory (LSTM) network. The transfer learning relies on a fixed-feature extraction. This novel deep learning approach successfully uses data of a low-frequency range, which is a precondition to use low-cost sensors. It is validated against the IEEE PHM 2012 Data Challenge, where it outperforms the winning approach. The results show its suitability for low-frequency sensor data and for efficient and effective transfer learning between different bearing types.

RVGAN-TL: A generative adversarial networks and transfer learning-based hybrid approach for imbalanced data classification

Imbalanced data distribution is the main reason for the performance degradation of most supervised classification algorithms. When dealing with imbalanced learning problems, the prediction of traditional classifiers tends to favor the majority class and ignore the minority class which is often much more important. Therefore, it is necessary to balance majority data and minority data before classification. A popular strategy for balancing the two data classes is synthesizing minority data. In recent years, generative adversarial networks (GAN) have shown great potential in fitting sample distributions. Based on this, this paper proposes a model combining improved GAN and transfer learning, RVGAN-TL, to solve the imbalanced learning problem of tabular data. As for the improvement of GAN, variational autoencoder (VAE) is used to generate latent variables with a posterior distribution as the input of GAN, and similarity measure loss is introduced into the generator to improve the quality of the minority data generated by GAN. In addition, a roulette wheel selection method is applied to the training data selection in GAN to rebalance data in the overlapping area. When data is balanced, the generated data is used as the source domain and the original data as the target domain, and the transfer learning method is used to train the final classifier. Experiments on 20 real datasets show that the classification performance of the proposed method is significantly improved compared with other popular methods.

A formulation-aid transfer learning-based framework in received power prediction

This study is motivated by the demand for an efficient deep learning-based model that helps us predict the future link quality for intelligent decision-making systems. In this letter, we propose a transfer learning-based approach to predict millimeter-wave future received power in an indoor environment. The model is pre-trained using formulation-aid generated data and fine-tuned using measured data. The proposed framework reduces more than 30% in root-mean-square error and 6.5% in accuracy with high training speed compared to the baseline training from scratch.

An Explainable AI-Enabled Framework for Interpreting Pulmonary Diseases from Chest Radiographs

Explainable Artificial Intelligence is a key component of artificially intelligent systems that aim to explain the classification results. The classification results explanation is essential for automatic disease diagnosis in healthcare. The human respiration system is badly affected by different chest pulmonary diseases. Automatic classification and explanation can be used to detect these lung diseases. In this paper, we introduced a CNN-based transfer learning-based approach for automatically explaining pulmonary diseases, i.e., edema, tuberculosis, nodules, and pneumonia from chest radiographs. Among these pulmonary diseases, pneumonia, which COVID-19 causes, is deadly; therefore, radiographs of COVID-19 are used for the explanation task. We used the ResNet50 neural network and trained the network on extensive training with the COVID-CT dataset and the COVIDNet dataset. The interpretable model LIME is used for the explanation of classification results. Lime highlights the input image's important features for generating the classification result. We evaluated the explanation using radiologists' highlighted images and identified that our model highlights and explains the same regions. We achieved improved classification results with our fine-tuned model with an accuracy of 93% and 97%, respectively. The analysis of our results indicates that this research not only improves the classification results but also provides an explanation of pulmonary diseases with advanced deep-learning methods. This research would assist radiologists with automatic disease detection and explanations, which are used to make clinical decisions and assist in diagnosing and treating pulmonary diseases in the early stage.

A Hybrid Deep Fused Learning Approach to Segregate Infectious Diseases

Humankind is facing another deadliest pandemic of all times in history, caused by COVID-19. Apart from this challenging pandemic, World Health Organization (WHO) considers tuberculosis (TB) as a preeminent infectious disease due to its high infection rate. Generally, both TB and COVID-19 severely affect the lungs, thus hardening the job of medical practitioners who can often misidentify these diseases in the current situation. Therefore, the time of need calls for an immediate and meticulous automatic diagnostic tool that can accurately discriminate both diseases. As one of the preliminary smart health systems that examine three clinical states (COVID-19, TB, and normal cases), this study proposes an amalgam of image filtering, data-augmentation technique, transfer learning-based approach, and advanced deep-learning classifiers to effectively segregate these diseases. It first employed a generative adversarial network (GAN) and Crimmins speckle removal filter on X-ray images to overcome the issue of limited data and noise. Each pre-processed image is then converted into red, green, and blue (RGB) and Commission Internationale de l’Elcairage (CIE) color spaces from which deep fused features are formed by extracting relevant features using DenseNet121 and ResNet50. Each feature extractor extracts 1000 most useful features which are then fused and finally fed to two variants of recurrent neural network (RNN) classifiers for precise discrimination of three-clinical states. Comparative analysis showed that the proposed Bi-directional long-short-term-memory (Bi-LSTM) model dominated the long-short-term-memory (LSTM) network by attaining an overall accuracy of 98.22% for the three-class classification task, whereas LSTM hardly achieved 94.22% accuracy on the test dataset.

A Transfer Learning-Based Approach to Fatigue Life Prediction of Corroded Bimetallic Steel Bars Using Small Samples

A Transfer Learning-Based Approach to Fatigue Life Prediction of Corroded Bimetallic Steel Bars Using Small Samples

Detection of COVID-19 Infection in CT and X-ray images using transfer learning approach.

The infection caused by the SARS-CoV-2 (COVID-19) pandemic is a threat to human lives. An early and accurate diagnosis is necessary for treatment. The study presents an efficient classification methodology for precise identification of infection caused by COVID-19 using CT and X-ray images. The depthwise separable convolution-based model of MobileNet V2 was exploited for feature extraction. The features of infection were supplied to the SVM classifier for training which produced accurate classification results. The accuracies for CT and X-ray images are 99.42% and 98.54% respectively. The MCC score was used to avoid any mislead caused by accuracy and F1 score as it is more mathematically balanced metric. The MCC scores obtained for CT and X-ray were 0.9852 and 0.9657, respectively. The Youden's index showed a significant improvement of more than 2% for both imaging techniques. The proposed transfer learning-based approach obtained the best results for all evaluation metrics and produced reliable results for the accurate identification of COVID-19 symptoms. This study can help in reducing the time in diagnosis of the infection.

A Deep Learning-Based Approach for the Diagnosis of Acute Lymphoblastic Leukemia

Leukemia is a deadly disease caused by the overproduction of immature white blood cells (WBS) in the bone marrow. If leukemia is detected at the initial stages, the chances of recovery are better. Typically, morphological analysis for the identification of acute lymphoblastic leukemia (ALL) is performed manually on blood cells by skilled medical personnel, which has several disadvantages, including a lack of medical personnel, sluggish analysis, and prediction that is dependent on the medical personnel’s expertise. Therefore, we proposed the Multi-Attention EfficientNetV2S and EfficientNetB3 state-of-the-art deep learning architectures using transfer learning-based fine-tuning approach to distinguish the normal and blast cells from microscopic blood smear images that both are pretrained on large-scale ImageNet database. We simply modified the last block of both models and added additional layers to both models. After including this Multi-Attention Mechanism, it not only reduces the model’s complexities but also generalizes its network quite well. By using the proposed technique, the accuracy has improved and the overall loss is also minimized. Our Multi-Attention EfficientNetV2S and EfficientNetB3 models achieved 99.73% and 99.25% accuracy, respectively. We have further compared the proposed model’s performance to other individual and ensemble models. Upon comparison, the proposed model outclassed the existing literature and other benchmark models, thus proving its efficiency.

Designing Hybrid CNN-SVM Model for COVID-19 Classification Based on X-ray Images Using LGBM Feature Selection

COVID-19 still exists at an alarming level; hence, early diagnosis is important for treating and controlling this disease due to its rapid spread. The use of X-rays in medical image analysis can play an essential role in fast and affordable diagnosis. This study used a two-level feature selection in hybrid deep convolutional features obtained from the extraction of X-ray images. The transfer learning-based approach was implemented using five convolutional neural networks (CNNs) named VGG16, VGG19, ResNet50, InceptionV3, and Xception. The combination of two or three CNNs' performance as a feature extractor was then carefully analyzed. We selected the features obtained from multiple CNNs in a particular layer with a specified percentage of features in the first level for getting relevant features from various models. Then, we combined those features and did the second level of feature selection to select the most informative features. Both levels of feature selection were carried out using the light gradient boosting machine (LightGBM) algorithm. The final feature set has been used to classify COVID-19 and non-COVID-19 chest X-ray images using the support vector machines (SVM) classifier. The proposed model's performance was evaluated and analyzed on the open-access dataset. The highest accuracy was 99.80% using only 5% of the features extracted from ResNet50 and Xception. The other way of combining the ensemble of deep features and a few recent works for the classification of COVID-19 were also compared with the proposed model. As a result, our proposed model has achieved the best success rate for this dataset and may be deployed to support decision systems for radiologists.

Quantitative analysis of metal scraps using laser-induced breakdown spectroscopy measurements via unsupervised adversarial domain adaptation

In this study, we propose a domain adaptation-based scheme for the compositional analysis of metal scraps using laser-induced breakdown spectroscopy (LIBS) measurements. LIBS analysis along with machine learning models has demonstrated feasibility for estimating elemental compositions of various materials such as metal scraps. However, LIBS-captured spectra can significantly vary depending on the physical and chemical properties of samples, configuration of the experimental setup, and prevailing environmental conditions such as temperature or humidity levels. Consequently, schemes based on calibration methods and the formulation of experimental parameters have been previously proposed by assuming that the training set is acquired under the same setup and conditions as the test set. Furthermore, current transfer learning solutions simply limit the pretrained model on the new test set, thus requiring a labeled test set in advance for finetuning. To address these limitations, this study proposes a novel transfer learning-based approach that depends on the domain adaptation of a source domain to transfer knowledge and reduce the difference between domains, and thus improving the performance of the target learner. We used LIBS measurements from certified reference materials (CRMs) to pretrain a source model to estimate the elemental concentration with known compositions of CRM metal samples. Then, the pretrained model is adapted in an unsupervised manner for the test/target dataset, and the adapted target model is used to quantify the elemental concentration of metal scraps. For evaluation, the CRMs of five representative metal types, including aluminum, copper, iron, stainless steel, and brass of known compositions, are measured under a typical LIBS setup to develop the pretrained model. For unsupervised domain adaptation and testing, metal scraps from actual industrial fields are tested using LIBS under different conditions such as lasers of different wavelengths and moving stages. The experimental results demonstrate that the proposed scheme achieves a root mean square error of 1.47 wt% or less for all test datasets, which is significantly better than the results of conventional regression schemes. The proposed quantitative analysis approach can be generalized to the regression of metal scraps and other industrial applications using LIBS.

VisionCervix: Papanicolaou cervical smears classification using novel CNN-Vision ensemble approach

About half a million women in the world are affected by cervical cancer and about 0.3 million deaths occur per year due to cervical cancer. Cytologists perform Pap-smear tests to screen the Pap Smear images of the cervical cells. This manual screening is prone also to error. Therefore, an automated computer-aided detection systems have been proposed for the classification of cervical cancer cell images. In the proposed work, an ensemble of Vision Transformer network (ViT) and convolution neural network (CNN) has been proposed for the classification of cervical cell Pap smear images. ViT has been known for its minimal inductive bias and its competitive classification performance in comparison to the state-of-the-art convolution neural network. Fine-tuning large ViT network is a computationally intensive procedure; therefore, as an alternative to ViT-CNN approach, another transfer learning-based approach has also been proposed in which the features extracted from the pre-trained CNNs are combined and classified with the resource-efficient Long Short Term Memory (LSTM) network. Comparison between both the approaches has been made on the basis of their classification performance, test time, generalization ability and attention maps. Experimental results show that the ViT-CNN ensemble approach achieved 97.65% classification accuracy whereas the LSTM-based approach achieved 95.80% classification accuracy. ViT-CNN ensemble approach achieves better classification accuracy at the cost of the huge demand for computation since it takes more computational resources in terms of high amount of random access memory (RAM) in the graphical processing unit (GPU); whereas, the CNN-LSTM approach is less accurate and computationally cheaper.

Deep Learning Utilization in Agriculture: Detection of Rice Plant Diseases Using an Improved CNN Model.

Rice is considered one the most important plants globally because it is a source of food for over half the world’s population. Like other plants, rice is susceptible to diseases that may affect the quantity and quality of produce. It sometimes results in anywhere between 20–40% crop loss production. Early detection of these diseases can positively affect the harvest, and thus farmers would have to be knowledgeable about the various disease and how to identify them visually. Even then, it is an impossible task for farmers to survey the vast farmlands on a daily basis. Even if this is possible, it becomes a costly task that will, in turn, increases the price of rice for consumers. Machine learning algorithms fitted to drone technology combined with the Internet of Things (IoT) can offer a solution to this problem. In this paper, we propose a Deep Convolutional Neural Network (DCNN) transfer learning-based approach for the accurate detection and classification of rice leaf disease. The modified proposed approach includes a modified VGG19-based transfer learning method. The proposed modified system can accurately detect and diagnose six distinct classes: healthy, narrow brown spot, leaf scald, leaf blast, brown spot, and bacterial leaf blight. The highest average accuracy is 96.08% using the non-normalized augmented dataset. The corresponding precision, recall, specificity, and F1-score were 0.9620, 0.9617, 0.9921, and 0.9616, respectively. The proposed modified approach achieved significantly better results compared with similar approaches using the same dataset or similar-size datasets reported in the extant literature.

Automatic identification of intestinal parasites in reptiles using microscopic stool images and convolutional neural networks.

Captive environments trigger the propagation and multiplication of parasites among different reptile species, thus weakening their immune response and causing infections and diseases. Technological advances of convolutional neural networks have opened a new field for detecting and classifying diseases which have shown great potential to overcome the shortcomings of manual detection performed by experts. Therefore, we propose an approach to identify six captive reptiles parasitic agents (Ophionyssus natricis, Blastocystis sp, Oxiurdo egg, Rhytidoides similis, Strongyloides, Taenia) or the absence of such parasites from a microscope stool images dataset. Towards this end, we first use an image segmentation stage to detect the parasite within the image, which combines the Contrast Limited Adaptive Histogram Equalization (CLAHE) technique, the OTSU binarization method, and morphological operations. Then, we carry out a classification stage through MobileNet CNN under a transfer learning scheme. This method was validated on a stool image dataset containing 3616 images data samples and 26 videos from the six parasites mentioned above. The results obtained indicate that our transfer learning-based approach can learn a helpful representation from the dataset. We obtained an average accuracy of 94.26% across the seven classes (i.e., six parasitic agents and the absence of parasites), which statistically outperformed, at a 95% confidence level, a custom CNN trained from scratch.

Transfer learning improves landslide susceptibility assessment

Landslide susceptibility assessment is often hindered by the lack of historical landslide records. In this study, we propose a transfer learning-based approach for landslide susceptibility assessment, aiming at substantially improving susceptibility prediction using knowledge outside the target domain, especially for regions with limited landslide data. The proposed method first trains a deep learning landslide susceptibility model (i.e., pre-trained model or source model) in a data-rich region (i.e., source domain). Transfer learning techniques are then applied to transfer the knowledge from the source domain to a new region (i.e., target domain) through model transfer and fine-tuning. The transferred model not only carries knowledge from the source domain but is also retrained with data from the target domain, hence achieving a much-improved performance in the new region even with very limited new data. A comprehensive case study in Hong Kong is conducted to investigate the feasibility of the proposed method and the influence of source domain scale on the transfer learning efficiency. Substantial improvements can be found with the proposed method: the accuracies on the test set of the target domain can be increased by 30% and the logarithmic losses can be decreased by 62%. We also reveal that transferring models from larger source domains can accomplish more improvements in both data-rich and data-limited cases. As the very first study that introduces deep transfer learning to landslide susceptibility assessment, the proposed method enables the sharing of landslide knowledge between regions, and is shown to be an intelligent and promising way for improving landslide susceptibility assessment for data-limited regions.