Transfer Learning Research Articles

Fire Recognition Method Based on PSO-BP Neural Network and ResNet50

The rapid development of modern society and continuous urbanization have resulted in a proliferation of functional buildings, which offer significant convenience to individuals, but pose significant fire hazards as well. How to detect the fire at the early stage is always the focus of research. This paper proposes a multi-information source fusion fire recognition method based on particle swarm optimization (PSO)-backpropagation (BP) neural networks and ResNet50. The PSO algorithm is applied to optimize the initial parameters of a BP neural network model, while data from three sensors — temperature, humidity and smoke — are integrated, through iterative training of the system, accurate recognition of sensor data can be achieved. Additionally, a method is proposed for the recognition of infrared fire images using ResNet50 and transfer learning. By improving the ResNet50 network model and migrating the ResNet50 pre-trained network weight, infrared fire image recognition accuracy is further enhanced. Then the sensor information recognition results and image information recognition results are input into the fuzzy system for fusion reasoning again, and the final decision is output according to the set fuzzy rules. Experimental findings demonstrate that the multi-information source fusion approach utilizing the PSO-BP neural network and ResNet50 significantly enhances the accuracy and response time of fire recognition, and achieves a remarkable recognition effect.

Malignancy Detection in Lung and Colon Histopathology Images by Transfer Learning with Class Selective Image Processing

Malignancy Detection in Lung and Colon Histopathology Images by Transfer Learning with Class Selective Image Processing

A Bearing Fault Diagnosis Model with Convolutional Cross Transformer and ResNet18

Abstract In the industrial field, malfunction of rotating machinery, especially bearings, can cause significant economic losses to enterprises. Addressing the limitations of traditional fault diagnosis methods, such as poor generalization performance and low noise resistance, this paper introduces a fault diagnosis model that parallels the Cross Convolutional Transformer and ResNet18 (CCTAR). The proposed CCTAR utilizes two feature extraction channels, aimed at balancing the extraction of local and global features, and the specially designed convolutional cross-decoding layer has excellent noise resistance, surpassing traditional multi-layer Transformer encoding layers with a single-layer structure. CCTAR achieves commendable recognition accuracy across multiple datasets and maintains high accuracy in noisy environments. Furthermore, transfer learning experiments have demonstrated the proposed model's capability to achieve superior fault diagnosis performance across different working conditions with a limited number of samples, highlighting its practical significance.

Rolling bearing fault diagnosis method based on dynamic simulated model from source domain to target domain with improved alternating transfer learning

Rolling bearing fault diagnosis method based on dynamic simulated model from source domain to target domain with improved alternating transfer learning

Advancements in Emotion Classification via Facial and Body Gesture Analysis: A Survey

ABSTRACTMachine vision has had a substantial impact on human‐computer interaction and psychological research. The ability of robots to generate realistic facial expressions has been enhanced by advancements in artificial intelligence, neural networks, and deep learning, thereby nurturing more profound emotive connections between humans and machines. This paper offers an in‐depth look of the most recent developments in emotion classification technologies, with a specific focus on the period from 2019 to 2023. This research addresses the integration of facial and body gesture analysis into emotion detection, emphasising the development of methods such as partial transfer learning, sign‐based measurement systems, and lightweight convolutional neural networks. By analysing the most recent research, the paper not only provides a comprehensive overview of the most advanced methods, but also emphasises their superiority over previous approaches in terms of efficiency, and applicability. The integration of gestures into emotion analysis is underscored as a critical area for further investigation, providing novel opportunities for comprehending and interpreting the intricate layers of nonverbal communication. The objective of this survey is to provide a resource for researchers and practitioners who are interested in advancing the field of emotion classification by utilising innovative methodologies.

Typical Scenario Load Identification Based on Feature Fusion and Transfer Learning

Background: The electricity demand is continuously increasing. However, various institutions, enterprises, and individuals exhibit many irregularities in their electricity usage, leading to significant wastage of electricity. To achieve effective energy management, researchers are attempting to analyze and regulate users' electricity demands by monitoring their load usage through Non- Intrusive Load Monitoring (NILM) technology. The accuracy of load identification in this technology will greatly impact the results of load monitoring. Although there are currently many articles and patents related to NILM, they utilize a large amount of computational resources and require high sampling rates from devices, yet the results are still unsatisfactory. Therefore, it is necessary to improve the accuracy of load identification in data with relatively low sampling frequencies. Objective: To improve the accuracy of load identification with low sampling frequency data, this paper proposes a typical scenario load identification method based on feature fusion and transfer learning. Methods: This method adopts the fusion of current and power factor angles to provide abundant identification information for NILM, effectively reducing the situation of single-feature overlap of different loads. By inputting the fused feature data into GoogLeNet and utilizing transfer learning for training, not only is the accuracy improved, but also the training time and the requirement for the sampling rate of training data are greatly reduced. In addition, selecting typical scenario loads can monitor loads in a targeted manner, reduce the waste of computing resources caused by irrelevant loads, and more effectively guide electricity usage strategies. Results: The proposed load identification method was tested on the low sampling frequency dataset used in this paper. It achieved an overall load identification accuracy of 94.61% across three scenarios, improving accuracy by 3% to 7% compared to other models. Conclusion: The simulation results indicate that this method achieves high load identification accuracy at low sampling frequencies. It also exhibits good generalization ability. This method not only reduces the performance requirements for monitoring equipment but also enhances monitoring efficiency.

Investigating Transfer Learning based CNN for Multi-Crop Disease Recognition in Horticulture Crops using Wireless Multimedia Sensor Network

Introduction: Achieving sustainable growth in agricultural productivity is crucial for addressing various challenges confronting food systems globally. One of the key aspects is effectively identifying and addressing diseases that can adversely affect crop quality and yield. Method: This research study proposes the use of Wireless Multimedia Sensor Network (WMSN) in conjunction with deep transfer learning model for disease detection in horticulture crops. The proposed system utilizes the image data of horticulture crops obtained using WMSN for performing a comparative assessment of several pre-trained deep learning models, encompassing VGG, ResNet, MobileNet, Inception, Xception, DenseNet, and EfficientNet, to ascertain their effectiveness in identifying diseases across a range of horticultural plants. The models undergo fine-tuning using transfer learning technique and are assessed on different measures, such as accuracy, precision, recall, F1 score, and the number of epochs needed for convergence using the image data obtained from WMSN. Results: Among the compared models, EfficientNetB3 appears as the best performing model, surpassing others in all evaluation metrics while also exhibiting low parameters and a compact size. Notably, EfficientNetB3 achieves a testing accuracy of 93.69% and a training accuracy of 98.85% after only 21 epochs of training. In contrast, DenseNet201, DenseNet169, and EfficientNetB0 demonstrate performance close to the best-performing models, but they have higher parameters or sizes. On the contrary, ResNet50, InceptionV3, and Xception exhibit poorer performance, presumably due to their substantial size in relation to the dataset. Conclusion: This analysis provides valuable insights for optimizing the top-performing models for disease classification across various plant species. Moreover, the fine-tuned EfficientNetB3 model shows potential for integration with mobile devices, facilitating real-time disease classification.

Machine learning analysis of high-repetition-rate two-dimensional Thomson scattering spectra from laser-produced plasmas

Abstract With the emergence of high-repetition-rate two-dimensional Thomson scattering (TS) measurements, improving spec- tral data analysis is a key area of interest. We present a new way to derive the electron temperature and density of laser-driven blast waves in plasmas from their TS spectra with machine learning (ML). This analysis occurs in both the non-collective (α < 1) and collective (α > 1) scattering regimes with the goal of autonomously and more accurately determining Te and ne both where spectral data has been collected and to give the ability to predict these attributes in regions where data has not been collected. We introduce three ML models, one trained only on experimental data, one only on synthetic data, and one using transfer learning, and compare their speed and accuracy with the conventional TS inversion algorithms in the open source PlasmaPy python package.

Deep learning techniques for diabetic retinopathy classification: a focus on VGG16 and EfficientNetB0

Diabetic Retinopathy (DR) affects millions worldwide, posing a serious eye condition that requires timely detection and diagnosis to prevent vision impairment and improve patient care. With the rise of Artificial Intelligence (AI), the medical field has gained powerful tools for early disease detection. This study explores the role of AI in the early diagnosis of DR, assessing the performance of two pre-trained convolutional neural networks (CNN) – VGG16 and EfficientNetB0. These models were fine-tuned and adapted using transfer learning techniques to classify DR and non-DR images. The evaluation was carried out using two distinct datasets from Kaggle, one containing RGB images and the other Gaussian-filtered images. The results demonstrated that VGG16, after fine-tuning, achieved an accuracy of 95.21%, while EfficientNetB0, employing transfer learning, reached 97.54% accuracy. These findings underscore the efficiency of both models in accurately identifying diabetic retinopathy.

Fast Iterative Sample Transfer Identification Method for Dynamic Systems Under Non‐identical Distribution

ABSTRACTThis paper proposes a method to improve the identification performance of linear dynamic systems by utilizing knowledge from samples of non‐identical distribution systems. Traditional identification methods heavily rely on the quality of the dataset, such as sample length and noise level, which constrains their performance due to the assumption of identical distribution. Motivated by the concept of sample‐based transfer learning, we propose a sample transfer identification method and derive the condition to avoid negative transfer. We develop a fast iterative transfer identification method for low storage costs, considering the computational burden imposed by the sample size from the source system. Additionally, based on the fast iterative transfer identification method, considering the need to update the current measurement data model in real time, a fast iterative online sample transfer identification method is explored. Through simulations, we validate the effectiveness and superiority of the proposed methods. The results show that sample transfer identification is superior to non‐transfer identification and fast iterative sample transfer identification effectively reduces the calculation amount when dealing with low quality measurement data.

From COVID-19 to monkeypox: a novel predictive model for emerging infectious diseases

The outbreak of emerging infectious diseases poses significant challenges to global public health. Accurate early forecasting is crucial for effective resource allocation and emergency response planning. This study aims to develop a comprehensive predictive model for emerging infectious diseases, integrating the blending framework, transfer learning, incremental learning, and the biological feature Rt to increase prediction accuracy and practicality. By transferring features from a COVID-19 dataset to a monkeypox dataset and introducing dynamically updated incremental learning techniques, the model's predictive capability in data-scarce scenarios was significantly improved. The research findings demonstrate that the blending framework performs exceptionally well in short-term (7-day) predictions. Furthermore, the combination of transfer learning and incremental learning techniques significantly enhanced the adaptability and precision, with a 91.41% improvement in the RMSE and an 89.13% improvement in the MAE. In particular, the inclusion of the Rt feature enabled the model to more accurately reflect the dynamics of disease spread, further improving the RMSE by 1.91% and the MAE by 2.17%. This study underscores the significant application potential of multimodel fusion and real-time data updates in infectious disease prediction, offering new theoretical perspectives and technical support. This research not only enriches the theoretical foundation of infectious disease prediction models but also provides reliable technical support for public health emergency responses. Future research should continue to explore integrating data from multiple sources and enhancing model generalization capabilities to further enhance the practicality and reliability of predictive tools.

Epitope-anchored contrastive transfer learning for paired CD8+ T cell receptor–antigen recognition

Epitope-anchored contrastive transfer learning for paired CD8+ T cell receptor–antigen recognition

SGTL-SUIE: semantic attention-guided transfer learning method for stylization underwater image enhancement

SGTL-SUIE: semantic attention-guided transfer learning method for stylization underwater image enhancement

Deep Learning‐Based Variational Autoencoder for Classification of Quantum and Classical States of Light

AbstractAdvancements in optical quantum technologies have been enabled by the generation, manipulation, and characterization of light, with identification based on its photon statistics. However, characterizing light and its sources through single photon measurements requires efficient detectors and longer measurement times to obtain high‐quality photon statistics. Here, a deep learning‐based variational autoencoder (VAE) method is introduced for classifying single photon added coherent state (SPACS), single photon added thermal state (SPATS), and mixed states between coherent and SPACS as well as between thermal and SPATS of light. The semi‐supervised learning‐based VAE efficiently maps the photon statistics features of light to a lower dimension, enabling quasi‐instantaneous classification with low average photon counts. The proposed VAE method is robust and maintains classification accuracy in the presence of losses inherent in an experiment, such as finite collection efficiency, non‐unity quantum efficiency, finite number of detectors, etc. Additionally, leveraging the transfer learning capabilities of VAE enables successful classification of data of any quality using a single trained model. It is envisioned that such a deep learning methodology will enable better classification of quantum light and light sources even in poor detection.

LWFDTL: lightweight fusion deep transfer learning for oral Squamous cell Carcinoma diagnosis using Histopathological oral Mucosa

LWFDTL: lightweight fusion deep transfer learning for oral Squamous cell Carcinoma diagnosis using Histopathological oral Mucosa

Towards Self-Conscious AI Using Deep ImageNet Models: Application for Blood Cell Classification

The exceptional performance of ImageNet competition winners in image classification has led AI researchers to repurpose these models for a whole range of tasks using transfer learning (TL). TL has been hailed for boosting performance, shortening learning time and reducing computational effort. Despite these benefits, issues such as data sparsity and the misrepresentation of classes can diminish these gains, occasionally leading to misleading TL accuracy scores. This research explores the innovative concept of endowing ImageNet models with a self-awareness that enables them to recognize their own accumulated knowledge and experience. Such self-awareness is expected to improve their adaptability in various domains. We conduct a case study using two different datasets, PBC and BCCD, which focus on blood cell classification. The PBC dataset provides high-resolution images with abundant data, while the BCCD dataset is hindered by limited data and inferior image quality. To compensate for these discrepancies, we use data augmentation for BCCD and undersampling for both datasets to achieve balance. Subsequent pre-processing generates datasets of different size and quality, all geared towards blood cell classification. We extend conventional evaluation tools with novel metrics—“accuracy difference” and “loss difference”—to detect overfitting or underfitting and evaluate their utility as potential indicators for learning behavior and promoting the self-confidence of ImageNet models. Our results show that these metrics effectively track learning progress and improve the reliability and overall performance of ImageNet models in new applications. This study highlights the transformative potential of turning ImageNet models into self-aware entities that significantly improve their robustness and efficiency in various AI tasks. This groundbreaking approach opens new perspectives for increasing the effectiveness of transfer learning in real-world AI implementations.

TL-CStrans Net: a vision robot for table tennis player action recognition driven via CS-Transformer

Currently, the application of robotics technology in sports training and competitions is rapidly increasing. Traditional methods mainly rely on image or video data, neglecting the effective utilization of textual information. To address this issue, we propose: TL-CStrans Net: A vision robot for table tennis player action recognition driven via CS-Transformer. This is a multimodal approach that combines CS-Transformer, CLIP, and transfer learning techniques to effectively integrate visual and textual information. Firstly, we employ the CS-Transformer model as the neural computing backbone. By utilizing the CS-Transformer, we can effectively process visual information extracted from table tennis game scenes, enabling accurate stroke recognition. Then, we introduce the CLIP model, which combines computer vision and natural language processing. CLIP allows us to jointly learn representations of images and text, thereby aligning the visual and textual modalities. Finally, to reduce training and computational requirements, we leverage pre-trained CS-Transformer and CLIP models through transfer learning, which have already acquired knowledge from relevant domains, and apply them to table tennis stroke recognition tasks. Experimental results demonstrate the outstanding performance of TL-CStrans Net in table tennis stroke recognition. Our research is of significant importance in promoting the application of multimodal robotics technology in the field of sports and bridging the gap between neural computing, computer vision, and neuroscience.

Deep transfer learning in machinery remaining useful life prediction: A systematic review

Abstract As a novel paradigm in machine learning, deep transfer learning (DTL) can harness the strengths of deep learning for feature representation, while also capitalizing on the advantages of transfer learning for knowledge transfer. Hence, DTL can effectively enhance the robustness and applicability of the data-driven RUL prediction methods, and has garnered extensive development and research attention in machinery RUL prediction. Although there are numerous systematic review articles published on the topic of the DTL-based approaches, a comprehensive overview of the application of DTL in the RUL prediction for different mechanical equipment has yet to be systematically conducted. Therefore, it is imperative to further review the pertinent literature on DTL-based approaches. This will facilitate researchers in comprehending the latest technological advancements and devising efficient solutions to address the cross-domain RUL prediction challenge. In this review, a brief overview of the theoretical background of DTL and its application in RUL prediction tasks are provided at first. Then, a detailed discussion of the primary DTL methods and their recent advancements in cross-domain RUL prediction is presented. Next, the practical application of the current research is discussed in relation to the research object and its open-source data. More importantly, several challenges and further trend are further presented to conclude this paper in the end. We have reason to hope this work can offer convenience and inspiration to researchers seeking to advance in the field of RUL prediction.

Segmentation of tunnel water leakage based on a lightweight DeepLabV3+ model

Abstract The accurate and efficient detection of water leakage with complex backgrounds is crucial for the safety of metro operations. A lightweight segmentation method for metro tunnel water leakage based on transfer learning is proposed. Firstly, this is based on the Deeplabv3+ model and adopts MobileNetv3-Large as the backbone feature extraction network, which significantly reduces the network parameters and improves the detection speed; secondly, it incorporates the efficient channel attention mechanism, which enables the model to adaptively adjust the weights of the channel features and capture the inter-channel relationships in the image, which significantly improves the model’s ability for feature extraction ability; furthermore, for the problem of severe imbalance between positive and negative samples in the dataset, the recognition accuracy of complex samples is increased by optimizing the loss function; finally, the training method of transfer learning is utilized to solve the problem of scarcity of water leakage dataset, and to improve the model’s accuracy and generalization ability. The results show that the model has more significant detection accuracy and segmentation speed advantages than today’s mainstream semantic segmentation model. With strong generalization ability in complex environments (e.g., low illumination and multiple obstructions), the model can be used for intelligent operation and maintenance in metro tunnel projects.

Predictive pretrained transformer (PPT) for real-time battery health diagnostics

Modeling and forecasting the evolution of battery systems involve complex interactions across physical, chemical, and electrochemical processes, influenced by diverse usage demands and dynamic operational patterns. In this study, we developed a predictive pre-trained Transformer (PPT) model equipped with 1,871,114 parameters that enhance identification of both short-term and long-term patterns in time-series data. This is achieved through the integration of convolutional layers and probabilistic sparse self-attention mechanisms, which collectively enhance prediction accuracy and efficiency in diagnosing battery health. Moreover, the customized hybrid-model fusion supports parallel computing and employs transfer learning, reducing computational costs while enhancing scalability and adaptability. Consequently, this allows for precise real-time health estimations across various battery cycles. We validated this method using a public dataset of 203 commercial lithium iron phosphate (LFP)/graphite batteries charged at rates ranging from 1C to 8C. By using only partial charge data—from an 80 % state of charge to the maximum charging voltage (3.6 V for LFP batteries, 4.2 V for ternary batteries)—and avoiding complex feature engineering, error metrics were achieved below 0.3 % for root mean square error (RMSE), weighted mean absolute percentage error (WMAPE), and mean absolute error (MAE), with an R2 of 98.9 %. The generalization capabilities were further demonstrated across 36 different testing protocols, encompassing 23,480 cycles throughout the entire life cycle, with a total inference time of 9.88 s during the testing phases. Further experiments on 30 nickel cobalt aluminum (NCA) batteries and 36 nickel cobalt manganese (NCM) batteries, across different battery types and operational scenarios, resulted in RMSE, WMAPE, and MAE all below 0.9 %, with R2 values of 94.1 % and 94.4 %, respectively. These findings highlight the potential of our customized deep transfer neural networks to enhance diagnostic accuracy, accelerate training, and improve generalization in real-time applications.