Proposed Transfer Learning Method Research Articles

Intelligent Prediction of Rate of Penetration Using Mechanism-Data Fusion and Transfer Learning

Rate of penetration (ROP) is crucial for evaluating drilling efficiency, with accurate prediction essential for enhancing performance and optimizing parameters. In practice, complex and variable downhole environments pose significant challenges for mechanistic ROP equations, resulting in prediction difficulties and low accuracy. Recently, data-driven machine learning models have been widely applied to ROP prediction. However, these models often lack mechanistic constraints, limiting their performance to specific conditions and reducing their real-world applicability. Additionally, geological variability across wells further hinders the transferability of conventional intelligent models. Thus, combining mechanistic knowledge with intelligent models and enhancing model stability and transferability are key challenges in ROP prediction research. To address these challenges, this paper proposes a Mechanism-Data Fusion and Transfer Learning method to construct an intelligent prediction model for ROP, achieving accurate ROP predictions. A multilayer perceptron (MLP) was selected as the base model, and training was performed using data from neighboring wells and partial data from the target well. The Two-stage TrAdaBoost.R2 algorithm was employed to enhance model transferability. Additionally, drilling mechanistic knowledge was incorporated into the model’s loss function as a constraint to achieve a fusion of mechanistic knowledge and data-driven approaches. Using MAPE as the measure of accuracy, compared with conventional intelligent models, the proposed ROP prediction model improved prediction accuracy on the target well by 64.51%. The model transfer method proposed in this paper has a field test accuracy of 89.71% in an oilfield in China. These results demonstrate the effectiveness and feasibility of the proposed transfer learning method and mechanistic–data integration approach.

Transfer learning reconstructs submarine topography for global mid-ocean ridges

Mid-ocean ridges are unique, tectonically active geographical units on Earth that profoundly control the ocean environment and dynamics at the global scale. However, high-resolution topographic data from mid-ocean ridges are rarely available due to the difficulty in detecting ocean floors, which further limits ocean research at the global scale. Here, we divide the global mid-ocean ridge system into 2805 tiles and reconstruct their high-resolution topography by using a transfer learning approach with freely available low-resolution digital elevation models (DEMs) and limited high-resolution DEMs. A high-frequency terrain feature-based deep residual network is proposed to generate high-resolution global mid-ocean ridge DEMs. In this network, topographic knowledge related to mid-ocean ridges is integrated and quantified to improve the learning efficiency and reconstruction quality of the network. A series of verifications and evaluations demonstrate the reliability of reconstructed topographies for submarine topography research. We observe that reconstructed topography can achieve good environmental understanding and information acquisition in the global mid-ocean ridge range. We find that the complexity of the previous terrain environment is underestimated by 26.63% in terms of the slope gradient and by 14.95% in terms of terrain relief, while a 101.10% information improvement can be obtained for the reconstructed topography. The reconstructed topography indicates that diverse and intricate topographical environments of mid-ocean ridges exist among different ocean regions. The proposed transfer learning method for reconstructing high-resolution mid-ocean ridge topographies is valuable and can be utilized for reconstructing information in regions that are difficult to observe directly and lack sufficient data.

Enhancing capacity estimation of retired electric vehicle lithium-ion batteries through transfer learning from electrochemical impedance spectroscopy

The low economic feasibility caused by inefficient testing and inaccurate performance estimation is one of the main bottlenecks in the echelon utilization of large-scale retired batteries. This study proposes a fast and accurate capacity estimation method for retired batteries based on electrochemical impedance spectroscopy (EIS). Firstly, the EIS of the batteries that experience multi-condition aging in the laboratory are collected. EIS characteristic parameter sequences highly related to battery performance, including real part and magnitude, are directly extracted to establish a base bi-directional long short-term memory model. Secondly, a transfer learning method based on feature matching is designed, which applies a linear transformation layer to map the features between the source and target domains. The proposed transfer learning method has been effectively validated on laboratory battery data measured at different temperatures and retired battery datasets of different material types. The improvements are especially notable for retired batteries. The detection time has been reduced, with each cell requiring only 1.67 min. And using only a small amount of data as input for transfer learning can achieve an accuracy improvement of over 90 %, indicating an effective transfer channel from the base model established on laboratory small-capacity battery aging data to large-capacity retired battery data is successfully established for the first time. For retired nickel-cobalt-manganese batteries, the mean absolute percentage error (MAPE) and the root mean square percentage error (RMSPE) are 2.33 % and 2.75 %, respectively, while for retired lithium-iron-phosphate batteries, the MAPE and RMSPE reached 4.12 % and 5.04 %, respectively. The results demonstrate the proposed method reduces the cost of repeated testing, modeling, and training for specific retired batteries while maintaining the accuracy of capacity estimation. This advancement helps to improve the efficiency of large-scale retired battery grading, and injects new momentum into facilitating more effective decision-making processes.

GNN-Based Concentration Prediction With Variable Input Flow Rates for Microfluidic Mixers.

Recent years have witnessed significant advances brought by microfluidic biochips in automating biochemical protocols. Accurate preparation of fluid samples is an essential component of these protocols, where concentration prediction and generation are critical. Equipped with the advantages of convenient fabrication and control, microfluidic mixers demonstrate huge potential in sample preparation. Although finite element analysis (FEA) is the most commonly used simulation method for accurate concentration prediction of a given microfluidic mixer, it is time-consuming with poor scalability for large biochip sizes. Recently, machine learning models have been adopted in concentration prediction, with great potential in enhancing the efficiency over traditional FEA methods. However, the state-of-the-art machine learning-based method can only predict the concentration of mixers with fixed input flow rates and fixed sizes. In this paper, we propose a new concentration prediction method based on graph neural networks (GNNs), which can predict output concentrations for microfluidic mixters with variable input flow rates. Moreover, a transfer learning method is proposed to transfer the trained model to mixers of different sizes with reduced training data. Experimental results show that, for microfluidic mixers with fixed input flow rates, the proposed method obtains an average reduction of 88% in terms of prediction errors compared with the state-of-the-art method. For microfluidic mixers with variable input flow rates, the proposed method reduces the prediction error by 85% on average. Besides, the proposed transfer learning method reduces the training data by 84% for extending the pre-trained model for microfluidic mixers of different sizes with acceptable prediction error.

Deep learning-based dispersion prediction model for hazardous chemical leaks using transfer learning

Machine learning has been employed for the rapid and accurate prediction of gas leak dispersion. However, existing models require extensive high-precision numerical simulation datasets, which presents a significant computational challenge. This study proposes a novel method using transfer learning for real-time dispersion prediction of hazardous chemical leaks. A deep learning model is pre-trained using all samples of a specific chemical, followed by fine-tuning with a small subset of samples from another chemical. This process enables the transfer of dispersion models across different chemicals and achieves rapid prediction of dispersion consequences for various chemicals. Optimal model hyperparameters are determined through a genetic algorithm, while the optimal numbers of reused layers and training samples are explored through sensitivity analysis. The proposed transfer learning method significantly reduces numerical computations by 52–74% while maintaining high prediction accuracy. This approach can provide guidance for developing deep learning-based dispersion prediction models for hazardous chemical leaks and help with emergency response strategies.

Transferring a deep learning model from healthy subjects to stroke patients in a motor imagery brain–computer interface

Objective. Motor imagery (MI) brain–computer interfaces (BCIs) based on electroencephalogram (EEG) have been developed primarily for stroke rehabilitation, however, due to limited stroke data, current deep learning methods for cross-subject classification rely on healthy data. This study aims to assess the feasibility of applying MI-BCI models pre-trained using data from healthy individuals to detect MI in stroke patients. Approach. We introduce a new transfer learning approach where features from two-class MI data of healthy individuals are used to detect MI in stroke patients. We compare the results of the proposed method with those obtained from analyses within stroke data. Experiments were conducted using Deep ConvNet and state-of-the-art subject-specific machine learning MI classifiers, evaluated on OpenBMI two-class MI-EEG data from healthy subjects and two-class MI versus rest data from stroke patients. Main results. Results of our study indicate that through domain adaptation of a model pre-trained using healthy subjects’ data, an average MI detection accuracy of 71.15% ( ±12.46% ) can be achieved across 71 stroke patients. We demonstrate that the accuracy of the pre-trained model increased by 18.15% after transfer learning (p < 0.001). Additionally, the proposed transfer learning method outperforms the subject-specific results achieved by Deep ConvNet and FBCSP, with significant enhancements of 7.64% (p < 0.001) and 5.55% (p < 0.001) in performance, respectively. Notably, the healthy-to-stroke transfer learning approach achieved similar performance to stroke-to-stroke transfer learning, with no significant difference (p > 0.05). Explainable AI analyses using transfer models determined channel relevance patterns that indicate contributions from the bilateral motor, frontal, and parietal regions of the cortex towards MI detection in stroke patients. Significance. Transfer learning from healthy to stroke can enhance the clinical use of BCI algorithms by overcoming the challenge of insufficient clinical data for optimal training.

The Temperature Field Prediction and Estimation of Ti-Al Alloy Twin-Wire Plasma Arc Additive Manufacturing Using a One-Dimensional Convolution Neural Network

Plasma arc deposition as an additive manufacturing technology has unique advantages for producing parts with complex shapes through layer-by-layer deposition. It is critical to predict and control the temperature field during the production process due to the temperature distribution and gradients determining the properties and performance of the part. Numerical simulation approaches, such as the finite element method, which provides a large amount of data for machine learning modeling, thus reducing the overhead of experimental measurements, are widely used in machine learning. In this paper, we propose a neural network combined finite element method and process prediction workflow. A one-dimensional convolutional neural network model for predicting 2D temperature distribution is developed by training the collected data on the planar temperature field of titanium–aluminum twin-wire plasma arc additive manufacturing and the finite element method. The results show that the predicted temperature mean square error is only 0.5, with less than a 20 °C error in peak temperature and a relative error below 1%. The proposed transfer learning method achieves the same training loss and is 500 iterations faster than basic training, which improves the training speed by 25%. The current study confirms the accurate performance of the ML model and the effectiveness of the optimization method.

Riemannian transfer learning based on log-Euclidean metric for EEG classification.

Brain computer interfaces (BCI), which establish a direct interaction between the brain and the external device bypassing peripheral nerves, is one of the hot research areas. How to effectively convert brain intentions into instructions for controlling external devices in real-time remains a key issue that needs to be addressed in brain computer interfaces. The Riemannian geometry-based methods have achieved competitive results in decoding EEG signals. However, current Riemannian classifiers tend to overlook changes in data distribution, resulting in degenerated classification performance in cross-session and/or cross subject scenarios. This paper proposes a brain signal decoding method based on Riemannian transfer learning, fully considering the drift of the data distribution. Two Riemannian transfer learning methods based log-Euclidean metric are developed, such that historical data (source domain) can be used to aid the training of the Riemannian decoder for the current task, or data from other subjects can be used to boost the training of the decoder for the target subject. The proposed methods were verified on BCI competition III, IIIa, and IV 2a datasets. Compared with the baseline that without transfer learning, the proposed algorithm demonstrates superior classification performance. In contrast to the Riemann transfer learning method based on the affine invariant Riemannian metric, the proposed method obtained comparable classification performance, but is much more computationally efficient. With the help of proposed transfer learning method, the Riemannian classifier obtained competitive performance to existing methods in the literature. More importantly, the transfer learning process is unsupervised and time-efficient, possessing potential for online learning scenarios.

An innovative generalization method for data-driven models of steering feedback torque

Accurately modeling steering feedback torque (SFT) is crucial for the performance of both steer-by-wire systems and driving simulators of vehicles. Physics-based methods have poor model accuracy and real-time performance, due to the model complexity and unknown or inaccurate model parameters. Therefore data-driven methods have gained increasing attention in recent years for their advantages on SFT modeling. However, developing satisfactory data-driven models requires a large amount of data with sufficient coverage on various driving conditions, which can be time-consuming and expensive. To address this issue, this paper proposes an innovative generalization method that can transfer an SFT model trained for a specific vehicle to other target vehicles, dramatically reducing the dependence on large amounts of data and the coverage of various driving conditions, thus, saving time and cost. First, a pre-trained SFT model with high accuracy and strong generalization capability is built based on large, full-coverage data from source domain vehicles. Then, a network is designed to fine-tune the pre-trained model with a small amount of data from target domain vehicles with only a few driving conditions. To identify the most suitable driving conditions for training the fine-tuning network, the performance of five networks trained by different driving conditions is compared. Finally, a control network with the same structure as the fine-tuned network is trained based on the same target domain data, and the model accuracy of the transfer network and the controlled network under different conditions is compared and analyzed. The proposed transfer learning method along with the designed transfer network is proved to be effective on improving prediction accuracy of the target domain by 58%–74.7% compared to the pre-trained network.

Deep learning mango fruits recognition based on tensorflow lite

Agricultural images such as fruits and vegetables have previously been recognised and classified using image analysis and computer vision techniques. Mangoes are currently being classified manually, whereby mango sellers must laboriously identify mangoes by hand. This is time-consuming and tedious. In this work, TensorFlow Lite was used as a transfer learning tool. Transfer learning is a fast approach in resolving classification problems effectively using small datasets. This work involves six categories, where four mango types are classified (Harum Manis, Langra, Dasheri and Sindhri), categories for other types of mangoes, and a non-mango category. Each category dataset comprises 100 images, and is split 70/30 between the training and testing set, respectively. This work was undertaken with a mobile-based application that can be used to distinguish various types of mangoes based on the proposed transfer learning method. The results obtained from the conducted experiment show that adopted transfer learning can achieve an accuracy of 95% for mango recognition. A preliminary user acceptance survey was also carried out to investigate the user’s requirements, the effectiveness of the proposed functionalities, and the ease of use of its proposed interfaces, with promising results.

Precise prediction of launch speed for athletes in the aerials event of freestyle skiing based on deep transfer learning

Automatically obtaining the launch speed are powerful guarantees for athletes in the aerials event of freestyle skiing to achieve good results. In most of the published studies describing athletes getting high scores, the assisting sliding distance depends entirely on the coach and even the athlete’s own experience, which may not be optimal. The main goal of the present paper is to use an acquisition system and develop an artificial neural network (ANN) model to automatically obtain the corresponding relationship between assisting sliding distance and speed. The influence of snow friction coefficient, wind speed, wind direction, slope, height and weight can be simulated in the Unity3D engine. The influence of temperature, humidity and tilt angle needs to be measured in real world by professional testers which is strenuous. The neural network is first trained by sufficient simulation data to obtain the encoded feature. Then, the information learned in simulation environment is transferred to another network. The second network uses the data taken from twenty professional testers. Compared with the model without transfer learning, the performance of proposed method has significant improvement. The mean squared error for the testing set is 0.692. It is observed that the speed predicted by the designed deep transfer learning (DTL) model is in good agreement with the experimental measurement results. The results indicate that the proposed transfer learning method is an efficient model to be used as a tool for predicting the assisting sliding distance and launch speed for athletes in the aerials event of freestyle skiing.

Short-term load forecasting based on AM-CIF-LSTM method adopting transfer learning

Aiming at the unreliability of historical data for short-term load forecasting caused by the sudden change of power grid load under emergencies, a short-term load prediction method adopting transfer learning is studied. The proposed transfer learning method combines the attention mechanism (AM) with the long short-term memory network coupled with input and forgetting gates (CIF-LSTM) to construct the AM-CIF-LSTM short-term load prediction model. First, the variational modal decomposition (VMD) method is used to extract the trend component and certain periodic high-frequency components of the load datasets of the scene to be predicted and similar scenes. Subsequently, the AM-encoder/decoder learning model is established based on the trend component, and the AM learnable parameters are trained and transferred to the AM-CIF-LSTM model. Furthermore, inspired by the idea of classified forecasting, the load trend component and periodic high-frequency components under the required prediction scene are predicted by AM-CIF-LSTM and deep recursive neural network (DRNN), respectively. Finally, the load forecasting results are superimposed to obtain the load forecasting value. The experimental results demonstrate that the proposed method outperformed the existing methods in multiple accuracy indicators and could predict the rapid change trend of load in the case of insufficient data accurately and stably.

SmartFPS: Neural network based wireless-inertial fusion positioning system.

Current wireless-inertial fusion positioning systems widely adopt empirical propagation models of wireless signals and filtering algorithms such as the Kalman filter or the particle filter. However, empirical models of system and noise usually have lower accuracy in a practical positioning scenario. The biases of predetermined parameters would enlarge the positioning error through layers of systems. Instead of dealing with empirical models, this paper proposes a fusion positioning system based on an end-to-end neural network, along with a transfer learning strategy for improving the performance of neural network models for samples with different distributions. Verified by Bluetooth-inertial positioning in a whole floor scenario, the mean positioning error of the fusion network was 0.506 m. The proposed transfer learning method improved the accuracy of the step length and rotation angle of different pedestrians by 53.3%, the Bluetooth positioning accuracy of various devices by 33.4%, and the average positioning error of the fusion system by 31.6%. The results showed that our proposed methods outperformed filter-based methods in challenging indoor environments.

A Glaucoma Detection System Based on Generative Adversarial Network and Incremental Learning

Among various eye diseases, glaucoma is one of the leading causes of blindness. Glaucoma is also one of the most common eye diseases in Taiwan. Glaucoma screenings can use optical coherence tomography (OCT) to locate areas in which the retinal nerve fiber layer is thinning. However, because OCT equipment is costly, only large hospitals with well-equipped facilities will have OCT, and regular eye clinics cannot afford such expensive equipment. This has caused many glaucoma patients to worsen because they cannot get an early diagnosis in regular eye clinics in time. This paper proposes a method of using a generative adversarial network (GAN) to generate corresponding OCT images from fundus images to assist family doctors in judging whether further examination is needed based on the generated OCT images to achieve early detection and treatment of glaucoma. In addition, in order to improve the classification accuracy of the system deployed in different hospitals or clinics, this paper also proposes to use the incremental training method to fine-tune the model. The model can be quickly applied by adding a small number of images from a specific clinic or hospital. Experimental results show that the cosine similarity between the generated OCT image and the real OCT image is 97.8%. Combined with the proposed transfer learning method, the classification accuracy of the classification model reaches 83.17%. As well as the use of the incremental method, the accuracy of identifying glaucoma is approximately 78.94%, which is 8.77% higher than the 70.17% accuracy of the initial model. Experimental results show the effectiveness and feasibility of our proposed method.

A novel heterogeneous transfer learning method based on data stitching for the sequential coding brain computer interface

ObjectiveFor the brain computer interface (BCI), it is necessary to collect enough electroencephalography (EEG) signals to train the classification model. When the operation dimension of BCI is large, it will bring great burden to data acquisition. Fortunately, this problem can be solved by our proposed transfer learning method. MethodFor the sequential coding experimental paradigm, the multi-band data stitching with label alignment and tangent space mapping (MDSLATSM) algorithm is proposed as a novel heterogeneous transfer learning method. After filtering by multi-band filtering, the artificial signals can be obtained by data stitching from the source domain, which build a bridge between the source domain and target domain. To make the distribution of two domains closer, their covariance matrices are aligned by label alignment. After mapping to the tangent space, the features are extracted from the Riemannian manifold. Finally, the classification results are obtained with feature selection and classification. ResultsOur data set includes the EEG signals from 16 subjects. For the heterogeneous transfer learning of cross-label, the average classification accuracy is 78.28%. MDSLATSM is also tested for cross-subject, and the average classification accuracy is 64.01%, which is better than existing methods. SignificanceCombining multi-band filtering, data stitching, label alignment and tangent space mapping, a novel heterogeneous transfer learning method can be achieved with superior performance, which promotes the practical application of the BCI systems.

Transfer Learning Approach and Nucleus Segmentation with MedCLNet Colon Cancer Database.

Machine learning has been recently used especially in the medical field. In the diagnosis of serious diseases such as cancer, deep learning techniques can be used to reduce the workload of experts and to produce quick solutions. The nuclei found in the histopathology dataset are an essential parameter in disease detection. The nucleus segmentation was performed using the colorectal histology MNIST dataset for nucleus detection in this study. The graph theory, PSO, watershed, and random walker algorithms were used for the segmentation process. In addition, we present the 10-class MedCLNet visual dataset consisting of the NCT-CRC-HE-100K dataset, LC25000 dataset, and GlaS dataset that can be used in transfer learning studies from deep learning techniques. The study proposes a transfer learning technique using the MedCLNet database. Deep neural networks pre-trained with the proposed transfer learning method were used in the classification with the colorectal histology MNIST dataset in the experimental process. DenseNet201, DenseNet169, InceptionResNetV2, InceptionV3, ResNet152V2, ResNet101V2, and Xception deep learning algorithms were used in transfer learning and the classification studies. The proposed approach was analyzed before and after transfer learning with different methods (DenseNet169 + SVM, DenseNet169 + GRU). In the performance measurement, using the colorectal histology MNIST dataset, 94.29% accuracy was obtained in the DenseNet169 model, which was initiated with random weights in the multi-classification study, and 95.00% accuracy after transfer learning was applied. In comparison with the results obtained from empirical studies, it was demonstrated that the proposed method produced satisfactory outcomes. The application is expected to provide a secondary evaluation for physicians in colon cancer detection and the segmentation.

Transfer Learning for Flow Reconstruction Based on Multifidelity Data

Reduced-order modeling for multifidelity flow reconstruction offers increased accuracy while saving cost in data generation. The key to obtaining successful multifidelity models lies in properly capturing the correlation between low-fidelity and high-fidelity data. In this work, we propose to apply transfer learning to discover representative features that better correlate the multifidelity data to improve the accuracy in multifidelity flow reconstruction. Essentially, transfer learning facilitates the modeling task through transferring the knowledge learned in a related task, thus improving model performance in multifidelity problems. In particular, a typical class of transfer learning based on domain adaptation is introduced to discover domain-invariant features from the flow data of multiple sources. Two transfer learning frameworks via either the transfer component analysis or the geodesic flow kernel are established, providing different concepts to match transferred features between multifidelity data. Thereafter, the transferred features can be used as inputs to build the bridge function between low-fidelity and high-fidelity data for flow reconstruction. The proposed transfer learning methods are tested by various cases with increasing complexity, including heat convection, the Burgers equation, and transonic flows past a NACA0012 airfoil and an ONERA M6 wing. Advantages of the proposed transfer learning method to obtain better feature representations and to help construct more accurate multifidelity models for flow reconstruction are presented.

A weighted adaptive transfer learning for tool tip dynamics prediction of different machine tools

In the batch machining of manufacturing enterprises, there are always different machine tools of the same type used. Due to the influence of the uncertain degree of deterioration during the service period and other factors, there are great differences between different machine tools of the same type. These factors lead to different tool tip dynamics even for the same tool-tool holder assembly under the same conditions, thus affecting the prediction of stability in the milling process. In this paper, a weighted adaptive joint distribution adaptation transfer learning method was proposed to predict the position-speed dependent tool tip dynamics of different machine tools that had different service time. Firstly, a machine tool was selected as the source machine tool, and the corresponding tool tip dynamics were identified as the source data through different position-speed milling tests. Then, for the new same type of machine tools, that was, the target machine tools, only a few different position-speed combinations milling tests need to be carried out for the same tool, so as to identify the tool tip dynamics as the target data. Subsequently, a Kriging regression model for predicting the position-speed dependent tool tip dynamics of the target machine tool was trained by the proposed transfer learning method. Compared with the prediction results of the current transfer learning method, it was found that the average test errors of the natural frequency and damping ratio of the proposed method were minimum, which were 0.48% and 7.2% respectively. Finally, the accuracy and effectiveness of the proposed method were verified by the actual milling experiments.

Not All Electrode Channels Are Needed: Knowledge Transfer From Only Stimulated Brain Regions for EEG Emotion Recognition.

Emotion recognition from affective brain-computer interfaces (aBCI) has garnered a lot of attention in human-computer interactions. Electroencephalographic (EEG) signals collected and stored in one database have been mostly used due to their ability to detect brain activities in real time and their reliability. Nevertheless, large EEG individual differences occur amongst subjects making it impossible for models to share information across. New labeled data is collected and trained separately for new subjects which costs a lot of time. Also, during EEG data collection across databases, different stimulation is introduced to subjects. Audio-visual stimulation (AVS) is commonly used in studying the emotional responses of subjects. In this article, we propose a brain region aware domain adaptation (BRADA) algorithm to treat features from auditory and visual brain regions differently, which effectively tackle subject-to-subject variations and mitigate distribution mismatch across databases. BRADA is a new framework that works with the existing transfer learning method. We apply BRADA to both cross-subject and cross-database settings. The experimental results indicate that our proposed transfer learning method can improve valence-arousal emotion recognition tasks.

Endoscopy image enhancement method by generalized imaging defect models based adversarial training

Objective. Smoke, uneven lighting, and color deviation are common issues in endoscopic surgery, which have increased the risk of surgery and even lead to failure. Approach. In this study, we present a new physics model driven semi-supervised learning framework for high-quality pixel-wise endoscopic image enhancement, which is generalizable for smoke removal, light adjustment, and color correction. To improve the authenticity of the generated images, and thereby improve the network performance, we integrated specific physical imaging defect models with the CycleGAN framework. No ground-truth data in pairs are required. In addition, we propose a transfer learning framework to address the data scarcity in several endoscope enhancement tasks and improve the network performance. Main results. Qualitative and quantitative studies reveal that the proposed network outperforms the state-of-the-art image enhancement methods. In particular, the proposed method performs much better than the original CycleGAN, for example, the structural similarity improved from 0.7925 to 0.8648, feature similarity for color images from 0.8917 to 0.9283, and quaternion structural similarity from 0.8097 to 0.8800 in the smoke removal task. Experimental results of the proposed transfer learning method also reveal its superior performance when trained with small datasets of target tasks. Significance. Experimental results on endoscopic images prove the effectiveness of the proposed network in smoke removal, light adjustment, and color correction, showing excellent clinical usefulness.