Residual attention-based tracking-by-detection network with attention-driven data augmentation

Tracking-by-detection (TBD) is a significant framework for visual object tracking. However, current trackers are usually updated online based on random sampling with a probability distribution. The performance of the learning-based TBD trackers is limited by the lack of discriminative features, especially when the background is full of semantic distractors. We propose an attention-driven data augmentation method, in which a residual attention mechanism is integrated into the TBD tracking network as supplementary references to identify discriminative image features. A mask generating network is used to simulate changes in target appearances to obtain positive samples, where attention information and image features are combined to identify discriminative features. In addition, we propose a method for mining hard negative samples, which searches for semantic distractors with the response of the attention module. The experiments on the OTB2015, UAV123, and LaSOT benchmarks show that this method achieves competitive performance in terms of accuracy and robustness.     

Numerical study on laminar burning velocity of ammonia flame with methanol addition

The combustion characteristics of ammonia/methanol mixtures were investigated numerically in this study. Methanol has a dramatic promotive effect on the laminar burning velocity (LBV) of ammonia. Three mechanisms from literature and another four self-developed mechanisms constructed in this study were evaluated using the measured laminar burning velocities of ammonia/methanol mixtures from Wang et al. (Combust.Flame. 2021). Generally, none of the selected mechanisms can precisely predict the measured laminar burning velocities at all conditions. Aiming to develop a simplified and reliable mechanism for ammonia/methanol mixtures, the constructed mechanism utilized NUI Galway mechanism (Combust.Flame. 2016) as methanol sub-mechanism and the Otomo mechanism (Int. J. Hydrogen. Energy. 2018) as ammonia sub-mechanism was optimized and reduced. The reduced mechanism entitled ‘DNO-NH3’, can accurately reproduce the measured laminar burning velocities of ammonia/methanol mixtures under all conditions. A reaction path analysis of the ammonia/methanol mixtures based on the DNO-NH3 mechanism shows that methanol is not directly involved in ammonia oxidation, instead, the produced methyl radicals from methanol oxidization contribute to the dehydrogenation of ammonia. Besides, NO_x emission analysis demonstrates that 60% methanol addition results in the highest NO_x emissions. The most important reactions dominating the NO_x consumption and production are identified in this study.     

Mechanistic implications of the solvent kinetic isotope effect in the hydrolysis of NaBH4

The sodium borohydride, NaBH₄, hydrolysis mechanism is studied via the H₂O/D₂O kinetic isotope effect (KIE). This reaction is of importance as NaBH₄ is considered as a hydrogen storage material. Nowadays, hydrogen is thought to be one of the most promising and efficient clean energy carriers. In order to control the rate of the hydrogen evolution reaction (HER), one has to understand the mechanism of its production. The H₂O/D₂O KIE of the reactions of NaBH₄ and NaBD₄ with water was studied in solutions containing a ratio of H₂O/D₂O = 1.00. The separation factor, α, of both reactions is α = 5.0 ± 1.0. The rate of the hydrolysis of BD₄^? in H₂O is faster than that of BH₄^?. The results point out that the rate-determining step in all hydrolysis stages is the H–OH bond scission.     

Sequential alignment attention model for scene text recognition

Scene text recognition has been a hot research topic in computer vision due to its various applications. The state-of-the-art solutions usually depend on the attention-based encoder-decoder framework that learns the mapping between input images and output sequences in a purely data-driven way. Unfortunately, there often exists severe misalignment between feature areas and text labels in real-world scenarios. To address this problem, this paper proposes a sequential alignment attention model to enhance the alignment between input images and output character sequences. In this model, an attention gated recurrent unit (AGRU) is first devised to distinguish the text and background regions, and further extract the localized features focusing on sequential text regions. Furthermore, CTC guided decoding strategy is integrated into the popular attention-based decoder, which not only helps to boost the convergence of the training but also enhances the well-aligned sequence recognition. Extensive experiments on various benchmarks, including the IIIT5k, SVT, and ICDAR datasets, show that our method substantially outperforms the state-of-the-art methods.     

Training visual attention in a naturalistic visual environment

The efficiency of training visual attention in the central and peripheral visual field was investigated by means of a visual detection task that was performed in a naturalistic visual environment including numerous, time-varying visual distractors. We investigated the minimum number of repetitions of the training required to obtain the top performance and whether intra-day training improved performance as efficiently as inter-day training. Additionally, our research aimed to find out whether exposure to a demanding task such as a microsurgical intervention may cancel out the effects of training.

Results showed that performance in visual attention peaked within three (for tasks in the central visual field) to seven (for tasks in the periphery) days subsequent to training. Intra-day training had no significant effect on performance. When attention training was administered after exposure to stress, improvement of attentional performance was more pronounced than when training was completed before the exposure. Our findings support the implementation of training in situ at work for more efficient results.

Practitioner Summary: Visual attention is important in an increasing number of workplaces, such as with surveillance, inspection, or driving. This study shows that it is possible to train visual attention efficiently within three to seven days. Because our study was executed in a naturalistic environment, training results are more likely to reflect the effects in the real workplace.     

Investigation on elemental mercury removal by cerium modified semi-coke

Coal-fired power plant is the largest anthropogenic mercury source. Active carbon injection technique has been widely used to control the mercury emissions. However, high operation cost limits its development and it is necessary to find other potential mercury sorbents. In this study, raw semi-coke and a series of novel cerium (Ce) modified semi-cokes were synthesized and utilized for removing elemental mercury (Hg⁰) from simulated flue gas. It is noteworthy that the efficiencies were tested without hydrogen chloride (HCl) in order to evaluate the sorbents efficacy for low-chlorine (Cl) coal. The results show that the modified sorbents exhibited the best performance at 150 °C. The performance of sorbent could be reinforced due to the existence of oxygen (O₂), nitric oxide (NO) and HCl. The adverse effect caused by sulfur dioxide (SO₂) reduced dramatically after Ce modification. The negative impact of ammonia (NH₃) on Hg⁰ removal in this study could be neglected owing to the tiny concentration of NH₃. Raw semi-coke provided sufficient carbon content, which is favorable to mercury adsorption. As Ce loading increased, the carbon structure changed and the crystal of cerium oxide was formed in the modified semi-coke. The mass fraction of cerium oxide on the sorbent was over 4.4% when the concentration of Ce modification solution was higher than 0.2 mol L⁻¹. The redox reaction activity and the oxygen storage ability of Ce³⁺/Ce⁴⁺ gave a huge boost to the performance of modified semi-coke. The addition of Ce also had an impact on the proportion of oxygen species.     

Deep multi-scale adversarial network with attention: A novel domain adaptation method for intelligent fault diagnosis

Data driven-based intelligent fault diagnosis methods, as a promising approach, have been widely employed in the health management and maintenance decision of rotating machinery. However, the domain shift phenomenon caused by internal and external interference inevitably exists in practical application scenarios, which significantly deteriorates the performances of the intelligent diagnosis model. And the preparation of label information in real complex scenes is usually time-consuming and expensive. To overcome these challenges, a novel unsupervised domain adaptation framework named deep multi-scale adversarial network with attention (MSANA) is introduced for machinery fault diagnosis. It is established based on two main components, one is the shared feature generator, which is constructed by two novel multi-scale modules with attention mechanism, and the other part is a fault pattern recognition module composed of two differentiated discriminators. While the multi-scale module is used to obtain rich features through different internal perceptual scales, the attention mechanism determines the weights of different scales, which promotes the dynamic adjustment performance and adaptive ability of the model. Then, decision boundary assisted adversarial learning strategy is employed to eliminate domain distribution differences and obtain domain-invariant features. A total of ten rolling bearing-based transfer scenarios and six gearbox-based transfer scenarios are adopted to evaluate the transferability of the proposed MSANA model, and the cross-domain transfer results show that it has superior transferability and stability.