基于非文本模态强化和门控融合方法的多模态情感分析

针对各模态之间信息密度存在差距和融合过程中可能会丢失部分情感信息等问题,提出一种基于非文本模态强化和门控融合方法的多模态情感分析模型。该模型通过设计一个音频-视觉强化模块来实现音频和视觉模态的信息增强,从而减小与文本模态的信息差距。之后,通过跨模态注意力和门控融合方法,使得模型充分学习到多模态情感信息和原始情感信息,从而增强模型的表达能力。在对齐和非对齐的CMU-MOSEI数据集上的实验结果表明,所提模型是有效的,相比现有的一些模型取得了更好的性能。

Dominant variation modes associated with Yangtze–Huai River Basin summer heavy rainfall events

The Yangtze–Huai River Basin(YHRB)always suffers from anomalously heavy rainfall during the warm season,and has been well explored as a whole area during the past several decades.In this study,the YHRB is divided into two core regions-the northern YHRB(nYHRB)and southern YHRB(sYHRB)-based on 29-year(1979–2007)June–July–August(JJA)temporally averaged daily rainfall rates and the standard deviation of rainfall.A spectral analysis of JJA daily rainfall data over these 29 years reveals that a 3–7-day synoptic-timescale high-frequency mode is absolutely dominant over the nYHRB,with 10–20-day and 15–40-day modes playing a secondary role.By contrast,3–7-day and 10–20-day modes are both significant over the sYHRB,with 7–14-day,15–40-day,and 20–60-day modes playing secondary roles.Based on a comparison between bandpass-filtered rainfall anomalies and original rainfall series,a total of 42,1,5,and 3 heavy rainfall events(daily rainfall amounts in the top 5%of rainy days)are detected over the nYHRB,corresponding to 3–7-day,7–14-day,10–20-day,and 15–40-day variation disturbances.Meanwhile,a total of 28,8,12,and 6 heavy rainfall events are detected over the sYHRB,corresponding to 3–7-day,7–14-day,10–20-day,and 20–60-day variation disturbances.The results have important implications for understanding the duration of summer heavy rainfall events over both regions.

The Static Stress Triggering Effects Related to the Yutian M_S7.3 Earthquake

This paper calculates the static Coulomb stress changes generated by four earthquakes in the Yutian area during 2008 ~ 2014 separately,then discusses the triggering influence,their accumulated Coulomb stress changes and their influence on nearby faults.The results indicate that the M S5.5 earthquake in 2011 and the M_S7.3 earthquake in 2014 are both in the regions where the Coulomb stress change is positive,the stress changes are 0.004 MPa and 0.021 MPa, respectively, meaning they are triggered by prior earthquakes.The M S6.2 earthquake in 2012 occurred in the place where Coulomb stress change was negative,so it is postponed by the prior earthquakes.The image of Coulomb stress changes of the M S7.3 earthquake in 2014 is in accord with aftershocks( M L≥ 3.0)distribution,but some regions on the fault where the Coulomb stress change is positive have few aftershocks,and strong aftershocks may occur at these districts in future.In addition,this paper calculates the Coulomb stress change on nearby faults,and finds that the Coulomb stress changes of different elements in the GGC fault are very different,and must receive strong triggered-influence,though the result may be influenced by the input finite fault model,so there is still a large earthquake-risk.The GGN,PLC,PLW and LBW faults were also triggered by the four earthquakes occurring between 2008 ~ 2014.Their maximum Coulomb stress changes all exceed 0.002 MPa,so they also have a strong earthquake hazard.