In this article, a deep learning (DL) method based on autoencoder network is proposed to achieve the inverse design of phase retrieval for large-scale antenna arrays. The inverse problem between the beam pattern and a...
详细信息
In this article, a deep learning (DL) method based on autoencoder network is proposed to achieve the inverse design of phase retrieval for large-scale antenna arrays. The inverse problem between the beam pattern and antenna phases is established first in the context of planar phased array. Inception-Resnet-V2 with prior knowledge (IR-PK) is proposed as an efficient model, which involves the prior knowledge of array factor to guide neural network (NN) learning for stronger fitting ability. To obtain the real-time phase retrieval in small terminals, a MobileNet-distilled IR-PK (MD-IR-PK) model combining lightweight architecture and knowledge distillation (KD) is then designed under the condition of limited resources. The method is validated for array beamforming and hologram. Compared with popular solutions, IR-PK shows the advantages of good accuracy, fast convergence, and computational efficiency. Experiments have been carried out for metasurface-based holography, with the measured results agreeing well with the simulated ones. The proposed method is competitive for complex electromagnetic (EM) inverse problems involving high nonlinearity.
Hyperspectral unmixing using deep learning has received increasing attention as a technique for estimating endmember spectra and fractional abundances of land covers. Among these, autoencoder-based methods are the mos...
详细信息
Hyperspectral unmixing using deep learning has received increasing attention as a technique for estimating endmember spectra and fractional abundances of land covers. Among these, autoencoder-based methods are the most common unsupervised unmixing approaches in hyperspectral image analysis, primarily involving three key steps: feature extraction, feature transformation to abundance, and endmember estimation. However, these methods face several challenges: (1) feature extraction is not directly related to the unmixing task, (2) the performance evaluation of endmembers and abundances is primarily based on reconstruction error, which may not fully capture their accuracy or physical interpretability, and (3) the variability of endmembers is often ignored. To address these problems, we propose a multistage graph-based autoencoder unmixing method. Firstly, a graph-based feature extraction mechanism is employed to extract features that encompass both global and local information, and specifically oriented towards the unmixing task. Then, to obtain accurate abundances and endmembers, abundance sparsity and low-rank priors within superpixels are incorporated to guide model training. Finally, to address the issue of spectral variability, it is assumed, based on physical plausibility, that the measured spectral signature for a material is identical within a superpixel, and related across different superpixels. Experimental results on both synthetic and real datasets demonstrate that the proposed method incorporating features oriented towards the unmixing task, prior knowledge tailored to specific scenarios, as well as accounting for the variability and invariance of endmembers, leads to more precise estimations of abundance and endmembers. The codes will be available at https://***/donghua898/MGA-Net.
During the downhole microseismic monitoring for hydraulic fracturing, microseismic signals are constantly vulnerable to interference from different kinds of noise. Improving the signal-to-noise ratio of microseismic r...
详细信息
During the downhole microseismic monitoring for hydraulic fracturing, microseismic signals are constantly vulnerable to interference from different kinds of noise. Improving the signal-to-noise ratio of microseismic records is always beneficial for processing and interpreting microseismic data. Unlike traditional methods that often result in the loss of signal details, an improved attention mechanism is proposed that can effectively enhance feature extraction from microseismic data and accurately recover detailed components in this article. To denoise the noisy three-component microseismic record effectively, we design a denoising network model that combines a convolutional autoencoder with an improved attention mechanism. Using the attention network to assign weights, channels containing noise information are given lower weights and effectively suppressed. Conventional methods and deep learning methods for denoising rarely consider the influence of polarization characteristics. The method proposed in this paper leverages deep learning for denoising while simultaneously reducing the impact of polarization characteristics throughout the denoising process. Simulation experiments are conducted using waveform analysis, time-frequency analysis, first arrival picking, and polarization analysis methods to validate the effectiveness of the model. Comparing the popular bidirectional long and short-term neural network, our model demonstrates superior recovery capabilities under various signal-to-noise ratio conditions.
Autonomous vehicles heavily rely on various sensors to evaluate their surroundings and issue essential control commands. Nonetheless, these sensors are susceptible to false data injection and spoofing attacks, posing ...
详细信息
Autonomous vehicles heavily rely on various sensors to evaluate their surroundings and issue essential control commands. Nonetheless, these sensors are susceptible to false data injection and spoofing attacks, posing a significant security threat. In response, this paper proposes a channel-spatial-temporal attention-based autoencoder network to detect sensor spoofing attacks on autonomous vehicles. The innovative network utilizes the reconstruction error based on the autoencoder to detect abnormalities in input time series data collected from multiple sensors. The proposed model consists of a memory-augmented based spatial-attention block and PSE-Res2Net block-based encoder and decoder. PSE-Res2Net block initially adopts Res2Net module to generate a multi-scale feature graph and enhance multi-dimensional representation ability of neural network, then applies the PSENet module to capture location-aware channel information and channel-sensitive spatial information through the interaction of channel attention and spatial attention. Moreover, the memory-augmented based temporal-attention block is developed to integrate multi-scale features and aggregate global sequence information of sensor measurements. Experimental evaluations conducted on the comma2k19, KITTI, and CCSAD datasets illustrate the superiority of the proposed detection model over baseline technologies. It exhibits enhanced performance in terms of mPre, mRe, mF1 score, and showcases heightened resilience against noise and attacks.
Spectral unmixing is an important task in hyperspectral image processing for separating the mixed spectral data pertaining to various materials observed aiming at analyzing the material components in observed pixels. ...
详细信息
Spectral unmixing is an important task in hyperspectral image processing for separating the mixed spectral data pertaining to various materials observed aiming at analyzing the material components in observed pixels. Recently, nonlinear spectral unmixing has received particular attention in hyperspectral image processing, as there are many situations in which the linear mixture model may not be appropriate and could be advantageously replaced by a nonlinear one. Existing nonlinear unmixing approaches are often based on specific assumptions on the nonlinearity and can be less effective when used for scenes with unknown nonlinearity. This article presents an unsupervised nonlinear spectral unmixing method that addresses a general model that consists of a linear mixture part and an additive nonlinear mixture part. The structure of a deep autoencoder network, which has a clear physical interpretation, is specifically designed to achieve this purpose. Moreover, a convolutional neural network (CNN) is used to capture the spectral-spatial priors from hyperspectral data. Extensive experiments with synthetic and real data illustrate the generality and effectiveness of this scheme compared with state-of-the-art methods.
In this paper, an end-to-end driving decision-making model is proposed for intelligent vehicle, utilizing a Variational autoencoder (VAE) network and Deep Reinforcement Learning to address the challenges in complex an...
详细信息
In this paper, an end-to-end driving decision-making model is proposed for intelligent vehicle, utilizing a Variational autoencoder (VAE) network and Deep Reinforcement Learning to address the challenges in complex and dynamical driving environments. Firstly, the traffic environment image features are extracted by VAE network, which can effectively reduce the amount of data input and improve the learning efficiency. Secondly, the Soft Actor-Critic (SAC) algorithm is improved through the application of TD error value constraints, N-step learning, etc. Then driving risk field and rule constraints are introduced into the improve SAC algorithm. Based on the real-time driving risk field, the skipping frame method can enhance learning efficiency, and the rule constraints can reduce the dangerous actions in the output of the algorithm. In order to verify the effectiveness of the model, in the CARLA simulation platform the models of scenario and algorithm are established, and the simulations are carried out. The results show that using decision-making model built by the proposed algorithm, the average driving distance by the intelligent vehicle has been improved by 91.37%, the average reward value of the task has been increased by 132.04%, the average success rate of the task has been improved by 46.56%, the training time is also significantly reduced. It demonstrated that the proposed decision-making model provides a significant improvement in driving safety and learning efficiency.
Understanding electrical load profiles and detecting anomaly behaviors are important to the smart grid system. However, current load identification and anomaly analysis are based on static analysis, and less considera...
详细信息
Understanding electrical load profiles and detecting anomaly behaviors are important to the smart grid system. However, current load identification and anomaly analysis are based on static analysis, and less consideration is given to anomaly findings under load change conditions. This paper proposes a deep variational autoencoder network (DVAE) for load profiles, along with anomaly analysis services, and introduces auto-time series data updating strategies based on sliding window adjustment. DVAE can help reconstruct the load curve and measure the difference between the original and the newer curve, whose measurement indicators include reconstruction probability and Pearson similarity. Meanwhile, the design of the sliding window strategy updates the data and DVAE model in a time-series manner. Experiments were carried out based on datasets from the U.S. Department of Energy and from Southeast China. The results showed that the proposed services could result in a 5% improvement in the AUC value, which helps to identify the anomaly behavior.
Hyperspectral unmixing is an important task in hyperspectral applications. Its essence is to estimate the spectra (endmembers) and corresponding proportion (abundances) of pure substances. In this paper, we propose a ...
详细信息
ISBN:
(数字)9781665427920
ISBN:
(纸本)9781665427920
Hyperspectral unmixing is an important task in hyperspectral applications. Its essence is to estimate the spectra (endmembers) and corresponding proportion (abundances) of pure substances. In this paper, we propose a new hyperspectral unmixing method with autoencoder network in wavelet domain. Based on the sparsity of wavelet coefficients, the high frequency parts are truncated to provide more reliable spectral similarity. After that, the batch normalization and ReLU function are followed to construct the hidden layer. In terms of loss function, the l(2) and l(1) norm are added to low and high frequency coefficients to ensure the energy fidelity and enhance the sparsity, respectively. Moreover, the hidden layer is characterized by l(1/2) norm to model the sparse prior of abundance, and SAD is used to enhance the spectral similarity. A large number of experiments show that the proposed method is superior to the most advanced methods.
Blind hyperspectral unmixing is an important technique in hyperspectral image analysis, aiming at estimating endmembers and their respective fractional abundances. Consider the limitations of using the linear model, n...
详细信息
Blind hyperspectral unmixing is an important technique in hyperspectral image analysis, aiming at estimating endmembers and their respective fractional abundances. Consider the limitations of using the linear model, nonlinear unmixing methods have been studied under different model assumptions. However, existing nonlinear unmixing algorithms do not fully exploit spectral and spatial correlation information. This paper proposes a nonsymmetric autoencoder network to overcome this issue. The proposed scheme benefits from the universal modeling ability of deep neural networks and enables to learn the nonlinear relation from the data. Particularly, the long short-term memory network (LSTM) structure is included to capture spectral correlation information, and a spatial regularization is introduced to improve the spatial continuity of results. An attention mechanism is also used to further enhance the unmixing performance. Experiments with synthetic and real data are conducted to illustrate the effectiveness of the proposed method.
Hyperspectral images (HSIs) contain a large number of mixed pixels due to low spatial resolution, which poses great challenges to the analyses and applications of HSIs. In recent years, convolutional neural networks (...
详细信息
ISBN:
(纸本)9781665403696
Hyperspectral images (HSIs) contain a large number of mixed pixels due to low spatial resolution, which poses great challenges to the analyses and applications of HSIs. In recent years, convolutional neural networks (CNNs) have attained promising performance in HSI field. However, few CNN-based methods are proposed to solve the hyperspectral unmixing (HU) problem because of insufficient labeled samples. In this paper, we propose a novel unsupervised method, sparsity constrained convolutional autoencoder network (SC-CAE), for the HU problem. The data are preprocessed by principal component analysis (PCA) and then fed into the encoder network to obtain low dimensional representations. The decoder network is to reconstruct the original data from these low dimensional representations. Under the sparse constraint, the endmember matrix and the abundance matrix are obtained after many training epochs. The experiment results on synthetic dataset and real dataset show that our method has evident advantages compared with several state-of-the-art methods.
暂无评论