We present a novel stacked autoencoder framework for feature extraction to improve classification of hyperspectral image, leveraging graph regularization to address the shortcomings of classical autoencoder that mainl...
详细信息
ISBN:
(纸本)9781665403696
We present a novel stacked autoencoder framework for feature extraction to improve classification of hyperspectral image, leveraging graph regularization to address the shortcomings of classical autoencoder that mainly focuses on learning spectral features. In the proposed method, we firstly construct a graph to represent the spectral-spatial similarity between pixels in a hyperspectral image by measuring their spatial and spectral distances. And then the graph regularized autoencoder is learned to transform the original spectral signatures of pixels into a new feature space used for the downstream pixel classification or other tasks. Our feature extraction method can preserve the intrinsic spectral-spatial distribution in a hyperspectral image and obtain more discriminative and robust features. The experiments on pixel classification show the competitive performance compared with classical autoencoder based and manifold learning based feature extraction approaches.
Our proposal consists of developing two novel activation functions in time series anomaly detection, they have the capability to reduce the validation loss. The approach is based on a current activation function in De...
详细信息
ISBN:
(纸本)9781728195438
Our proposal consists of developing two novel activation functions in time series anomaly detection, they have the capability to reduce the validation loss. The approach is based on a current activation function in Deep Learning, a very intensive field studied over time, in order to find the most suitable activation in a neural network. In order to achieve this purpose. we used an LSTM (Long Short-Term Memory) autoencoder architecture, using these two novel functions to see the network's behavior through introducing them. The key point in our proposal is given by the learnable parameter, assuring more flexibility within the network in weights' updates, in fact, this property being more powerful than a predefined parameter that will bring a constraint due to its limit. We tested our proposal in comparison to other popular functions such as ReLU (Linear Rectifier Unit), hyperbolic tangent (tanh), Into activation function. Also, the novelty of this paper consists of taking into consideration of piecewise behavior of an activation function in order to increase the performance of a neural network in Deep Learning.
Satellites are intricate systems that consist of large number of interconnected devices. Consequently, it generates tremendous amount of telemetry data during its lifetime. If the system information of solar array tha...
详细信息
ISBN:
(纸本)9789881563804
Satellites are intricate systems that consist of large number of interconnected devices. Consequently, it generates tremendous amount of telemetry data during its lifetime. If the system information of solar array that contained in the telemetry data can be used to forecast the future state, then the malfunction alarm and in-orbit maintenance can be implemented. At present, research on solar arrays basically focuses on state monitoring and outlier detection, and almost no fault prediction is involved. In this paper, a data-driven faults prediction method for solar array based on long short-term memory (LSTM) networks and autoencoder (AE) is proposed. This novel fault prediction method works by creating a prediction model using the collected time series. The temporal features of sensor data are captured with LSTM networks, which perform well in extracting the features of time series. The predicted sequence is reconstructed by the identifying model, and the reconstruction error of the input sequence which is used as a health indicator to estimate the states of devices. The data used in verification experiment in this paper is the telemetry data of an orbiting satellite. The experiments show that the proposed method can detect the occurrence of failures several days in advance compared with the traditional method.
Convolutional neural networks (CNNs) have been prominent in most hyperspectral image (HSI) processing applications due to their advantages in extracting local information. Despite their success, the locality of the co...
详细信息
Convolutional neural networks (CNNs) have been prominent in most hyperspectral image (HSI) processing applications due to their advantages in extracting local information. Despite their success, the locality of the convolutional layers within CNNs results in heavyweight models and time-consuming defects. In this study, inspired by the excellent performance of transformers that are used for long-range representation learning in computer vision tasks, we built a lightweight vision transformer for HSI classification that can extract local and global information simultaneously, thereby facilitating accurate classification. Moreover, as traditional dimensionality reduction methods are limited in their linear representation ability, a three-dimensional convolutional autoencoder was adopted to capture the nonlinear characteristics between spectral bands. Based on the aforementioned three-dimensional convolutional autoencoder and lightweight vision transformer, we designed an HSI classification network, namely the "convolutional autoencoder meets lightweight vision transformer" (CAEVT). Finally, we validated the performance of the proposed CAEVT network using four widely used hyperspectral datasets. Our approach showed superiority, especially in the absence of sufficient labeled samples, which demonstrates the effectiveness and efficiency of the CAEVT network.
This paper proposes a combined network model and synchrophasor data-based approach to formulate a non-linear state estimator for the reconstruction of the three-phase voltages of the nodes (buses) in an electric power...
详细信息
This paper proposes a combined network model and synchrophasor data-based approach to formulate a non-linear state estimator for the reconstruction of the three-phase voltages of the nodes (buses) in an electric power network. The network model uses network topology with impartial admittance information to create a first estimate of any missing three-phase complex-valued synchrophasor voltage measurements. These first estimates along with synchrophasor measurements from other nodes in the network are used in a trained autoencoder neural network to further improve the three-phase voltage state estimation. Instead of standard voltage magnitude and voltage angle synchrophasor data, the proposed method uses real and imaginary parts of the three-phase voltages for state reconstruction to address discontinuities in voltage phase angle wrapping. The approach is illustrated on an example network and actual synchrophasor data measurements to validate performance. The illustrations show that the non-linear state estimator that combines the network model and synchrophasor data-based approach can outperform a non-linear estimator that uses synchrophasor data only. Copyright (C) 2021 The Authors.
We present a reduced-order model (ROM) based on Koopman autoencoder architecture for prediction of current density in kinetic plasma simulations. We apply the proposed method to forecast the time evolution of the curr...
详细信息
ISBN:
(纸本)9781733509626
We present a reduced-order model (ROM) based on Koopman autoencoder architecture for prediction of current density in kinetic plasma simulations. We apply the proposed method to forecast the time evolution of the current density derived from high-fidelity electromagnetic particle-in-cell (EMPIC) simulation based on the Maxwell-Vlasov system of equations. We test the Koopman autoencoder architecture for a two-dimensional oscillating electron beam inside a square cavity. Such analysis can be crucial in modeling of complex nonlinear plasma system as well as expediting time-consuming particle-in-cell simulations.
Wearable sensor based human activity recognition is a challenging problem due to difficulty in modeling spatial and temporal dependencies of sensor signals. Recognition models in closed-set assumption are forced to yi...
详细信息
ISBN:
(纸本)9783030757687;9783030757670
Wearable sensor based human activity recognition is a challenging problem due to difficulty in modeling spatial and temporal dependencies of sensor signals. Recognition models in closed-set assumption are forced to yield members of known activity classes as prediction. However, activity recognition models can encounter an unseen activity due to body-worn sensor malfunction or disability of the subject performing the activities. This problem can be addressed through modeling solution according to the assumption of open-set recognition. Hence, the proposed self attention based approach combines data hierarchically from different sensor placements across time to classify closed-set activities and it obtains notable performance improvement over state-of-the-art models on five publicly available datasets. The decoder in this autoencoder architecture incorporates self-attention based feature representations from encoder to detect unseen activity classes in open-set recognition setting. Furthermore, attention maps generated by the hierarchical model demonstrate explainable selection of features in activity recognition. We conduct extensive leave one subject out validation experiments that indicate significantly improved robustness to noise and subject specific variability in body-worn sensor signals. The source code is available at: ***/saif-mahmud/hierarchical-attention-HAR.
Temporal Convolutional autoencoders are used as feature extractors to project time series onto a latent space where similarity detection can be easily performed. This model can generate accurate descriptors of the tem...
详细信息
ISBN:
(纸本)9781665403696
Temporal Convolutional autoencoders are used as feature extractors to project time series onto a latent space where similarity detection can be easily performed. This model can generate accurate descriptors of the temporal profile of the input time-series. We apply this algorithm to PolSAR S1 uncoherent SAR time series where the model learns highly discriminative data representations. This reduction method is compared to others such as PCA or Temporal Averaging and is shown to outperform them when leveraging the learnt representation using K-Means clustering.
Depression is one of the most common mental health problems, which can lead to significant mental disorders and suicidal behavior. To diagnose depression levels, patients with depressive disorders are required to comp...
详细信息
ISBN:
(纸本)9781665438759
Depression is one of the most common mental health problems, which can lead to significant mental disorders and suicidal behavior. To diagnose depression levels, patients with depressive disorders are required to complete self-assessment questionnaires. However, many depressed patients are misdiagnosed in clinical practice due to patients' missing data. In this paper, we introduce, APD, a novel data-driven approach based on autoencoder to predict the missing responses accurately. Inspired by existing autoencoder-based recommender systems, our autoencoder is based on collaborative filtering, which estimates unobserved data by cooperation with other patients' responses. Experimental results show that the proposed autoencoder-based prediction system outperforms the averaging and the linear models. We demonstrate that this model can be used to predict patients' depression status with a low error of 2.85%.
Fault detection is one of the most challenging tasks in industrial applications, which aims at identifying the faulty condition deviating from the normal condition of the machine. In this work, a fault detection metho...
详细信息
ISBN:
(纸本)9781665433860
Fault detection is one of the most challenging tasks in industrial applications, which aims at identifying the faulty condition deviating from the normal condition of the machine. In this work, a fault detection method is proposed based on autoencoders and online sequential extreme learning machines (OS-ELM). The autoencoder is employed for high-level feature extraction from the monitoring signal and the OS-ELM is developed based on features extracted from signals of normal condition. The fault detection is performed based on i) the updating of OS-ELM using the newly collected data;U) the quantification of the model modification. The data collected under the faulty condition is expected to significantly modify the OS-ELM model. The proposed fault detection method is validated considering a benchmark bearing case study.
暂无评论