All updateable objects in working memory are updated whenever any of them are modified: Evidence from the memory updating paradigm.

In a series of experiments, participants were required to keep track of 1 or 2 working memory (WM) objects, having to update their values in 80% of the trials. Updating cost, defined as the difference between update and non-update trials, was larger when 2 objects were involved compared with when there was only 1 object was involved. This finding was interpreted as evidence that the updating process encompasses both objects in WM, even though only 1 of them is actually updated. This feature of WM updating is limited to objects defined as "updateable," throughout the trial sequence. The results are explained by the need to reprogram the phonological loop when updating or the need for desynchronization followed by resynchronization of WM contents. (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

基于近似高斯金字塔的视觉注意模型快速算法

利用输入图像的近似高斯金字塔,将经典的基于显著性的视觉注意模型改造为时空开销更小的版本,从而使其更加适合在嵌入式实时系统中实现.首先采用矩形窗口近似圆形窗口,矩形平均算子近似高斯卷积核;然后采用“先做行累加,再做列累加”的方法来实现矩形平均算子,并直接采样计算出各个特征通道的显著性分布图,该算法关于输入图像像素点个数具有线性时间复杂度;最后,还给出了在显著性分布图中抑制已提取区域显著性的快速算法.在Berkeley分割图像库上的实验结果表明,该方法极大地减小了系统实现的时空开销,且输出结果的误差在可接受范围内.提出的用矩形窗口近似圆形窗口,用矩形平均算子近似高斯卷积核的方法,还适用于其他需要在嵌入式实时系统中实现的图像处理问题.     

基于视角置信度和注意力的暴力行为识别

暴力行为容易出现遮挡情况, 识别准确率较低. 目前, 一些算法加入多视角视频输入来解决遮挡问题, 以等量权重将所有视角数据融合, 但是不同视角的视频因拍摄距离和遮挡情况本身就对识别存在差异性. 针对该问题, 本文提出一种基于视角置信度和注意力的暴力行为识别方法, 提高暴力识别的准确率. 本文将时序差分模块TDM的输入扩展成多视角, 将通道注意力机制运用在片段维度来增强TDM中跨段特征提取能力, 通过背景抑制方法突显移动目标的纹理特征并计算出每个视角图像的置信度, 引入双线性池化方法融合多视角视频特征, 根据视角置信度分配每个视角局部特征的权重. 本文在公开数据集CASIA-Action和自制数据集上进行了验证. 实验表明, 本文提出的视角置信度方法优于改进前的双线性池化方法, 暴力行为准确率相较于现有的行为识别方法取得了更好的效果.     

基于多流网络一致性的视频显著性检测

现有的视频显著性检测算法通常采用双流结构提取视频的时空线索,其中运动信息作为双流结构的一个分支,在显著物体发生剧烈或慢速移动时存在运动估计准确率低的问题,并且不合理的训练数据或方案使得权重偏向单个分支结构。提出一种基于多流网络一致性的视频显著性检测算法MSNC。设计并使用一种新的三重网络结构提取预选目标区域的颜色信息、时序信息和先验特征,通过先验特征补偿运动流的缺陷,并提高运动线索的利用率。采用多流一致性融合模型优化三流分支,得到不同特征的最佳融合方案。同时通过循环训练策略平衡三重网络的权重,以避免网络过度拟合单流分支,从而有效地提高运动估计和定位的准确率。在Davis数据集上的实验结果表明,相比PCSA、SSAV、MGA等算法,该算法的鲁棒性更优,其maxF和S-Measure值分别达到0.893和0.912,MAE仅为0.021。     

When does visual attention select all features of a distractor?

What happens after visual attention is allocated to an object? Although many theories of attention assume that all of its features are selected and processed, there has been little direct evidence that an irrelevant feature dimension of an attended nontarget is processed. In 5 experiments presented here, the authors used a singleton paradigm to investigate the effect of attention on nontarget objects. Participants made a speeded feature discrimination of a target for which the response was either compatible or incompatible with an irrelevant feature dimension of a distractor. The results show that the irrelevant distractor features were processed to the point that they interfered with the response to the target. The response compatibility effect was observed even when the location of the target or the distractor was invariant, although it was much weaker when both locations were invariant. These results demonstrate that in many circumstances, an attended distractor is completely selected and fully processed, and the complete processing of distractors depends on a number of factors, many of which are related to the strength of attention to the distractor. (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

Working memory capacity and the top-down control of visual search: Exploring the boundaries of "executive attention".

The executive attention theory of working memory capacity (WMC) proposes that measures of WMC broadly predict higher order cognitive abilities because they tap important and general attention capabilities (R. W. Engle & M. J. Kane, 2004). Previous research demonstrated WMC-related differences in attention tasks that required restraint of habitual responses or constraint of conscious focus. To further specify the executive attention construct, the present experiments sought boundary conditions of the WMC-attention relation. Three experiments correlated individual differences in WMC, as measured by complex span tasks, and executive control of visual search. In feature-absence search, conjunction search, and spatial configuration search, WMC was unrelated to search slopes, although they were large and reliably measured. Even in a search task designed to require the volitional movement of attention (J. M. Wolfe, G. A. Alvarez, & T. S. Horowitz, 2000), WMC was irrelevant to performance. Thus, WMC is not associated with all demanding or controlled attention processes, which poses problems for some general theories of WMC. (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

对气体分析仪现状与发展的初步探讨

本文较详细地介绍了氩、氧、水分等几种分析仪器 ,阐述了它们的工作原理、优点、缺点及使用中应注意的问题     

The Attentional Dynamics of Masked Detection.

A dichoptic masking procedure was used to test whether the mask-dependent cuing effects found in luminance detection by P. L. Smith (2000a) were due to integration masking or interruption masking. Attentional cuing enhanced detection sensitivity (d') when stimuli were backwardly masked with either dichoptic or monoptic masks, whereas no cuing effect was found with unmasked stimuli, implying the mask dependencies were due to interruption of stimulus processing in visual cortex by the mask. The effect is predicted by a gated diffusion process model in which masks interrupt stimulus processing and attention controls the flow of information to a sequential-sampling decision mechanism. The model correctly predicts different patterns of performance for detection and discrimination and cuing effects in simple reaction time. (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

Capturing Attention When Attention "Blinks".

Four experiments addressed the question of whether attention may be captured when the visual system is in the midst of an attentional blink (AB). Participants identified 2 target letters embedded among distractor letters in a rapid serial visual presentation sequence. In some trials, a square frame was inserted between the targets; as the only geometric object in the sequence, it constituted a singleton. Capture effects obtained when the AB was most severe and when it was over were compared. There were 3 main results. First, capture occurred even when the AB was crippling, suggesting that a singleton exogenously engaged attention even when processing of a previous target was continuing apace. Second, when the singleton contained the key target feature, capture effects were clearly manifest. Third, even when the singleton did not possess the key target feature, it still succeeded in capturing attention, although the effects were both feeble and fleeting. (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

基于全局自注意力的小麦图像识别

在实际应用场景下,通过图像识别的方式来识别小麦的病虫害具有极大的挑战性。与以往纯粹基于卷积神经网络（Convolutional Neural Network, CNN）的方法相比,将小麦图像转换成一系列视觉语言,并从全局视角进行小麦识别的方法是更可行和实用的。运用Convolutional Visual Transformers（CVT）来解决小麦识别分为2个环节。首先,利用2分支CNN生成的2种特征图来实现注意选择性融合(Attentional Selective Fusion, ASF)。ASF通过融合多个特征和全局-局部注意力来获取有区别的信息,并投射成一系列的视觉语言。其次,受Transformers在自然语言处理方面的成功启发,用全局自注意力来建模这些视觉语言之间的关系。将CVT与经典分类网络LeNet-5、ResNet-18、VGG-16、EfficientNet对比,识别率有所提升,同时该方法具有良好的泛化能力。