The global modeling capability of vision transformer (ViT) has been well proven in the field of hyperspectral image (HSI) classification. However, ViT does not have the excellent local feature extraction capability compared with the convolutional neural network (CNN). Therefore, early-stage convolutions are often used to enhance ViT’s local representation ability. However, directly applying convolutions on high-dimensional HSI data increases computational overhead. Moreover, recent researches have observed that ViT may suffer from high redundancy in capturing multi-head self attention (MHSA). To address the above issues, we propose a multi-scale super token transformer (MSSTT) model for HSIs classification. We use a divide-and-conquer strategy to extract local features and global dependencies of HSIs data at multiple granularities. Specifically, our proposed model incorporates two branches: a multi-scale convolution (MSConv) branch that utilizes various convolutional kernels to extract diverse local features, and a multi-scale super token attention (MSSTA) branch for capturing global features with low redundancy. Finally, comparative experimental results with advanced methods show that the proposed MSSTT possesses better classification performance. On Salinas, Pavia University, and Kennedy Space Center datasets, the overall accuracies (OAs) of our MSSTT are 98.47%, 98.47%, and 99.38%, respectively. Code will be released at https://github.com/zhangtaizheng/MSSTT.
6 November 2024 in IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
Hyperspectral image (HSI) classification is a challenging task in remote sensing applications, aiming to determine the category of each pixel by utilizing rich spectral and spatial information in HSI. Convolutional neural networks (CNNs) have been effective in processing HSI data by extracting local features, but they are deficient in capturing global contextual information. Recently, transformer has become proficient in attending to global information due to their self-attention mechanisms, yet they may fall short in capturing multiscale features of HSI. To address these limitations, a global–local multigranularity transformer (GLMGT) network is proposed for HSI classification. The GLMGT combines CNN with the transformer to comprehensively capture multigranularity spectral and spatial features across global and local scales. Specifically, we introduce a multigranularity spatial feature extraction block to extensively extract spatial information at different granularities, including multiscale local spatial features and global spatial features. In addition, we introduce a multigranularity spectral feature extraction block to fully leverage spectral information across different granularities. The validity of the proposed method is demonstrated through experimental validation using seven publicly available datasets, which include two Chinese satellite hyperspectral datasets (ZY1-02D Huanghekou and GF-5 Yancheng) and one UAV-based hyperspectral dataset.
5 October 2022 in Measurement Science and Technology
The subway sliding plug door system has been opened and closed frequently for a long time under variable working conditions, and multiple failures are prone to occurring and resulting in train shutdowns and even major safety accidents. Due to the complex physical mechanism of the sliding plug door system, it is difficult for a single signal to accurately represent the failure states of the sliding plug door system. Thus, identifying the multiple failure causes of the subway sliding plug door system has become a challenging problem. Aiming at the problem, an equipment multiple failure causes intelligent identification method based on an integrated strategy is proposed for the subway sliding plug door system under variable working condition. In the proposed method, firstly, the sensitive features that can reflect the degradation state of equipment are obtained by using the random forest to measure the importance of fetatures and sort them. Secondly, feature dimensionality is reduced by using t-distributed stochastic neighbor mbedding (TSNE) to map the screened high-dimensional features to low-dimensional space. Finally, the parameters of the extreme learning machine (ELM) are optimized by using the particle swarm optimization (PSO) algorithm to build a multiple failure causes classification model. The proposed method is verified by the 1:1 benchmark test data of the subway sliding plug door system. The results show that the proposed method has higher classification accuracy, faster calculation speed, and stronger generalization ability. The proposed method is an effective integrated strategy to identify multiple failure causes in the subway sliding plug door system and guide the health management and operational maintenance of the subway sliding plug door system.
22 February 2024 in Transactions of the Institute of Measurement and Control
Planetary gearbox is a key component in modern industry. A sudden failure may cause disastrous consequences. Thus, accurately acquiring the fault severity can be of importance. Diversity entropy emerges as a promising feature extraction tool for monitoring the health condition. However, the original diversity entropy has the defect that the data length of multiple time series will shorten at deep scales, resulting in unstable complexity estimation at high scale. To overcome this defect, a new feature extraction method has been proposed named refined composite multiscale diversity entropy (RCMDE). The proposed RCMDE method combines moving average windows under each scale factor and the refined state probability to improve the statistical reliability, which allows the diversity entropy to explore more refined fault information hidden at deeper scales. The simulation and experiment results proved that the proposed method has the highest diagnostic accuracy with the best stability in fault severity identification of planetary gearbox.
27 August 2024 in Measurement Science and Technology
The metal oxide semiconductor field effect transistor (MOSFET) is subjected to various stresses due to the external and internal operating environments, leading to ageing and failure of the device. With multiple degradation mechanisms, a single piece of information can no longer fully reflect the health state of MOSFETs, so how to use multi-source data to characterise the state of the device and predict the remaining useful life (RUL) is an issue worth exploring. To address this problem, a method for constructing health indicators (HI) as well as predicting RUL using multi-source data is proposed. In this method, firstly, the features are computed by selecting the appropriate ageing signal from the ageing mechanism. Secondly, the extracted features are filtered using Pearson’s algorithm to find the features that are strongly correlated with longevity. Then, the filtered features are merged by dimensionality reduction using the kernel principal component analysis algorithm and the HI is constructed from the reduced result. Finally, an unsupervised clustering algorithm is used to classify the states based on the data distribution in HI, and the filtered features are used as input to the grey wolf optimisation bidirectional long short-term memory neural network to predict the RUL of the device. In this paper, the proposed method is validated using data from the MOSFET Accelerated Aging Experiment at the NASA Ames Centre of Excellence for Prediction. The results show that the method is able to achieve good results in health state assessment and RUL prediction of MOSFETs. The proposed method is an effective and comprehensive strategy that is more helpful for the operation and maintenance of electronics.
25 November 2024 in Measurement Science and Technology
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 remaining useful life (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.