Search

In recent years, image perception such as enhancement, classification and object detection with deep learning has achieved significant successes. However, in real world under extreme conditions, the training of a deep learning model often yields low accuracy, low efficiency in feature extraction and generalizability, due to the inner uncourteous and uninterpretable characteristics. In this paper, a maximal-entropy-attention-based convolutional neural network (MEA-CNN) framework is proposed. A maximum entropy algorithm is first used for image feature pre-extraction. An attention mechanism is then proposed by combining the extracted features on original images. By applying the mechanism, the key areas of an image are enhanced, and noised area can be ignored. Afterward, the processed images are transferred into region convolutional neural network, which is a well-known pre-trained CNN model, for further feature learning and extraction. Finally, two real-world experiments on traffic sign recognition and road surface condition monitoring are designed. The results show that the proposed framework has high testing accuracy, with improvements of 17% and 2.9%, compared with some other existing methods. In addition, the features extracted by the model are more easily interpretable. [ABSTRACT FROM AUTHOR]

To guarantee the safe operation of machinery and reduce its maintenance costs, estimating its remaining useful life (RUL) is a crucial task. Hence, in this study, a multi-scale memory-enhanced prediction method is proposed to describe fully characteristics of the data. This method is based on a deep learning algorithm and is designed to estimate the RUL of aircraft engines. To handle the complex and multi-fault operating conditions with uncertain properties in RUL estimation, a hybrid model that combines a multi-scale deep convolutional neural network and long short-term memory is presented. Experimental verification was carried out with the Commercial Modular Aero-Propulsion System Simulation dataset from NASA. Compared with multi-scale deep convolutional and long short-term memory networks, the hybrid model performed more efficiently. Furthermore, compared with other state-of-the-art methods, the multi-scale memory-enhanced prediction method can achieve better prognostics, especially for equipment with multiple operating conditions and failure modes. [ABSTRACT FROM AUTHOR]

This study presents the structural, thermodynamic, electronic, and optical properties of the β-BeO phase of ZnO under negative pressure calculated via first-principles method. We analyzed the phase transition by the local density approximation, and we find that the structural phase transition from wurtzite B4 to metastable β-BeO phase occurred at − 10.2 GPa. We used the Debye model to calculate the thermodynamic properties. The results show that the bulk modulus decreases with increasing temperature or decreasing pressure whereas the heat capacity and thermal expansion coefficient increase with increasing temperature or decreasing pressure. We employed the hybrid functional (Becke-Lee-Young-Parr exchange–correlation functional) method to predict the electronic and optical properties. The band structure calculations confirmed that the β-BeO phase possesses an indirect wider band gap than that of B4 phase. Redshift in band gap energy and optical properties are observed with decreasing pressure. The dielectric function ε1(ω) calculations validate that the β-BeO phase exhibits metallic behaviour at around 16.3 eV, whereas optical constants exploration has confirmed its transparent nature. The present study proves that the β-BeO phase ZnO may be potentially applicable in next-generation electronics. Additionally, it is helpful in material sciences to understand and examine the surrounding pressure during a variety of mineral formation processes. [ABSTRACT FROM AUTHOR]

In recent years, person re-identification based on video has become a hot topic in the field of person re-identification. The self-attention mechanism can improve the ability of deep neural networks in computer vision tasks such as image classification, image segmentation and natural language processing tasks. In order to verify whether the self-attention can improve the performance or not in person re-identification tasks, this paper applies two self-attention mechanisms, non-local attention and recurrent criss-cross attention to person re-identification model, and experiments are conducted on Market-1501, DukeMTMC-reID and MSMT17 person re-identification datasets. The results show that the self-attention mechanism can improve the accuracy of the person re-identification model. The accuracy is higher when the self-attention module is inserted into the convolutional layers of the re-identification network. [ABSTRACT FROM AUTHOR]

Optimization problems on unknown low-dimensional structures given by high-dimensional data belong to the field of optimizations on manifolds. Though recent developments have advanced the theory of optimizations on manifolds considerably, when the unknown low-dimensional manifold is given in the form of a set of data in a high-dimensional space, a practical optimization method has yet to be developed. Here, we propose a neural network approach to these optimization problems. A neural network is used to approximate a neighborhood of a point, which will turn the computation of a next point in the searching process into a local constraint optimization problem. Our method ensures the convergence of the process. The proposed approach applies to optimizations on manifolds embedded into Euclidean spaces. Experimental results show that this approach can effectively solve optimization problems on unknown manifolds. The proposed method provides a useful tool to the field of study low-dimensional structures given by high-dimensional data. [ABSTRACT FROM AUTHOR]

Two-photon fluorescence (TPF) of olive oils is discovered and observed experimentally for the first time. Variations of the single-photon fluorescence (SPF) and TPF with the excitation wavelength are investigated for four different olive oils. The results show that fluorescence of the cosmetic olive oils (COO) is very weak and exhibits only one spectral peak around 490 nm. While for the ordinary edible oils (OEO) whether they are during their shelf life or not, their fluorescence spectra may exhibit multiple peak structures. The short-term natural expiration only slightly weakens TPF of OEO. Moreover, the excitation wavelength affects the OEO spectra considerably in terms of the spectral peak number, the spectral peak position, and spectral shapes. When the excitation wavelength decreases from 700 nm, the whole TPF of the OEO also decreases. Relatively, however, the short wave band will decrease and disappear more quickly. While for the SPF, the long wave band will decrease and disappear first. The optimal excitation wavelengths to make the TPF strongest are around 700 nm and 640 nm for OEOs and COO, respectively. And effects of temperature on SPF and TPF of extra virgin olive oil are also explored. This work may be of significance for its potential applications in TPF detection and two-photon laser. [ABSTRACT FROM AUTHOR]

A structural self-organized DBN (S-DBN) is proposed in this paper to improve the ability of feature learning in unsupervised training. In S-DBN, the strategy of dropout is designed for unsupervised learning to reduce inner cooperation between feature detectors. Then, the regularization-reinforced transfer function is put forward, in order to further reduce the insignificant weights, and to raise the abilities of feature learning and generation. The fast training method of contrastive divergence is designed, and backpropagation is used in supervised training. Finally, two experiments on regression and classification using MNIST show that S-DBN has better generation and faster convergence rate than other methods, in particular, in regression experiment, the proposed model beats traditional DBN by 1.50%; in image classification, the proposed model achieves smaller testing error in much less computing time. [ABSTRACT FROM AUTHOR]

Magnetic resonant wireless power transfer (MR-WPT) has become a significant method for powering the measurement units in landslide monitoring boreholes. This paper focuses on the effects of quality factor, coupling distance and load on the power transfer stability, sensitivity and performance of an MR-WPT system. Firstly, through a numerical model of an MR-WPT, the effect of quality factor and coupling distance on the output power, transfer efficiency and frequency splitting phenomenon of MR-WPT systems have been comparatively analyzed. The relationship between the load and the coupling distance region corresponding to optimal transfer performance has been studied. Results show that improving the quality factor of the coils is beneficial for the transfer performance and stability of WPT system. Furthermore, through the selection of the load, a coupling distance region with high transfer stability and an optimal tradeoff between power and efficiency can be obtained. Finally, the above theoretical simulation results have been verified by experimental results. [ABSTRACT FROM AUTHOR]

The spin states, elastic properties and seismic velocities of ferric iron-bearing post-perovskite MgSiO (pPv) with single oxygen vacancy [Mg(Si,Fe)O and Mg(Si,Fe)O] were calculated by first principles based on density functional theory. The effects of ferric iron and oxygen vacancy on seismic waves were studied for the host pPv subjected to a hydrostatic pressure. Calculations revealed a new spin transition from intermediate-spin to low-spin states with increasing pressure. As a result, the volume was reduced and the elastic constants were modified, producing a clear decrease in the seismic velocities of both compressive wave and shear wave due to the reduction of bulk modulus and shear modulus. The ferric iron and oxygen vacancy also had a minor effect on wave anisotropy. [ABSTRACT FROM AUTHOR]

Adsorption of single gold (Au) atom at three kinds of sites (hollow, bridge and top) on the hydroxylated β-cristobalite SiO (1 1 1) surface was studied using the first-principles calculations with general gradient approximation (GGA). The results of adsorption energies and density of electronic states (DOS) suggest that the hollow and bridge sites have the basically equal capability of binding Au, while the ability of the Top site is weaker. Two new energy levels emerge after the adsorption at all sites; in DOS of the Hollow configuration, one locates at −0.15 eV, composed of Au 5d and O 2p electronic states, another just crosses through the Fermi level, consisting of Au 6s, H 1s and O 2p. In addition, Mulliken population analyses indicate that electron transfer takes place between the Au atom and the surface H and O atoms in the Hollow and Bridge configurations, which can be used to interpret the adsorption of Au onto the positions. However, neither H nor O chemically bonds with Au atom. [ABSTRACT FROM AUTHOR]