Results 41 to 50 of about 1,128,392 (212)
智能科学与技术学报, 2021 近年来，基于自注意力机制的神经网络在计算机视觉任务中得到广泛的应用。随着智能交通系统的广泛应用，面对复杂多变的交通场景，车牌识别任务的难度不断提高，准确识别的需求更加迫切。因此提出一个基于自注意力的免矫正的车牌识别方法T-LPR。首先对图像进行切片和序列化，并使用3D卷积对切片序列进行特征提取，从而得到图像的嵌入向量序列。然后将嵌入向量序列输入基于Transformer Encoder的编码器中，学习各个嵌入向量之间的关系并输出最终的编码结果。最后使用分类器进行分类。在多个公共数据集上的实验结果表明 ...
曾淦雄, 柯逍
doaj
Conservation Biology, Volume 40, Issue 2, April 2026.Abstract There exists a growing suite of technologies that support significant and exciting progress in biodiversity conservation and research. Citizen scientist participation is common in this research and often focuses on data collection and labeling.
Joycelyn Longdon +5 more
wiley +1 more source
Gaoya dianqi, 2023 语音识别是变电站智能运检中关键的人机交互技术。然而，由于生产环境中存在使用专业术语多和噪声大的问题，传统的语音识别方法的效果受限。为此，文中提出了一种基于声音谱特征的语音识别方法。通过融合MFCC与CQT谱，形成一种基于声音谱的特征参数，通过对参数分布的估计，能够有效地降低语音信息中的噪声干扰。为提升语音识别性能，文中设计一个端到端的语音识别模型。该模型基于卷积神经网络（CNN），并融合了CTC和注意力机制。CNN网络能够有效地捕捉语音数据中的局部模式和结构信息 ...
高宝明 +5 more
doaj
Rare Metals, Volume 45, Issue 3, March 2026.ABSTRACT The evolution of the γ′−Ni3(Al, Ti) phase in superalloys is governed by a coupled coalescence‐ripening mechanism driven by elemental diffusion. In this study, thermal exposure experiments were performed at 750°C and 800°C for durations ranging from 100 to 20,000 h.
Zhaotian Wang +5 more
wiley +1 more source
人工智能学刊深度学习与生成式智能作为现代人工智能体系的核心构成，其理论演进与算法创新直接决定智能模型的表征能力与生成效能。当前主流模型在特征提取过程中仍面临多尺度信息耦合不足、全局与局部特征关联薄弱等问题，生成环节则存在语义约束缺失、模态对齐精度有限、可控性理论支撑不足等瓶颈。本文以深度学习特征表征机制与生成式智能内容生成机理为研究内核，构建多尺度特征融合与动态权重自适应分配的优化框架，同时设计双层注意力机制与语义一致性约束相结合的生成控制模型，从理论层面揭示特征空间与生成分布的内在映射关系 ...
J. Nie
semanticscholar +1 more source
Jianzhu Gangjiegou Jinzhan, 2002 本文对钢材于制作与营建过程中产生之残留应力进行了分析和解释,详细阐述了残留应力产生的机理和实用的检测方法,提出了残留应力的消除方法,以及于规划及设计中应注意之事项。
陈纯森
doaj
Rare Metals, Volume 45, Issue 3, March 2026.ABSTRACT High‐entropy alloys (HEAs) have attracted considerable interest from researchers owing to their tunable chemical compositions, exceptional structural stability, and promising catalytic properties. However, their large‐scale application is often hindered by complex manufacturing techniques and poor durability.
Sheng Guo +9 more
wiley +1 more source
安全科学与工程学报针对复合扰动背景下电压暂降成因识别过程中存在的特征表达不足、类别间差异不明显以及复杂工况下识别精度受限等问题,提出一种基于双流特征融合与注意力机制的电压暂降成因识别方法。该方法以原始电压信号和增强辅助信号构建双流输入结构,其中原始支路保留完整时序波形信息,辅助支路则通过CEEMDAN-FFT提取重构基波,并结合FFAM幅值调制策略对暂降区间进行特征增强,从而突出波形边沿变化和局部幅值突变特征。在特征提取阶段,采用InceptionTime作为主干网络提取多尺度时序特征 ...
义钦 韩
semanticscholar +1 more source
People and Nature, Volume 8, Issue 2, Page 270-282, February 2026.Abstract Chinese consumer demand for ivory threatens vulnerable African elephant populations, despite long‐standing efforts to combat the illegal wildlife trade. Behavioural science approaches have been proposed to address these systemic behaviours. However, their robust use remains uncommon in demand management interventions.
Molly R. C. Brown +3 more
wiley +1 more source
Rare Metals, Volume 45, Issue 2, February 2026.ABSTRACT TiFe‐based hydrogen storage alloys have garnered significant attention because of their cost‐effectiveness, high hydrogen capacity, and favorable hydrogen absorption/desorption characteristics. However, practical application was limited by activation challenges and sensitivity to environmental poisoning.
Yuehai Li +8 more
wiley +1 more source