Search

Event Argument Extraction aims to extract various arguments of a specific event and assign the roles to them. Currently, most of previous work focused on sentence-level task, ignoring the whole contexts of the event mentions. Obviously, the event mentions in a discourse always are around a specific topic and have interior relations, especially those intra-sentence or adjacent-sentence event mentions. Therefore, this paper proposes an event argument extraction graph model on discourse level. It adds various event mentions, entity mentions and partial semantic dependencies into a graph, which aggregates the discourse contexts by event information, entity information and semantic dependency. The experimental results on the ACE2005 Chinese corpus show that our proposed model outperforms several strong baselines.

A line feature matching method for high-resolution images was proposed to improve the low significant level of a point feature and to overcome the matching shortage between weak texture images.Firstly,the edges of images were extracted and tracked to fit straight-lines.All straight-lines were classified into two groups: long-lines and short-lines.The long-lines were matched based on the direction relationship primarily.Then,the relationship-descriptors of short-lines were constructed using the angle and the Euclidean distance between the short-line and long correspondences.Finally,short-lines were matched according to the similarity of their relationship-descriptors.The experimental results demonstrate that the proposed line matching algorithm is robust for the scale,rotation and illumination.As all the lines have corresponding linear equations in the same image,the image feature has higher significant level and can avoid the mismatching.The probability of correct matches of the algorithm exceeds 90% and its root mean square error has achieved sub-pixel level.The performance of the proposed algorithm is better than that of the point-based method,especially in weak texture areas.

The dust case was analyzed by using aerosol extinction coefficient sequence chart retrieved by lidar observation data and meteorological observation data.The key factor of these two dust cases occurred was the dust transferred from North China,and the key reason of those two dust cases maintain so long was exogenous input and the vertical wind.The atmospheric profile affected the intensity of dust case.The dust case was holding when the temperature inversion occurred at the ground.The solar radiation affected the height of boundary layer.The main particulate matter is big size particle in the first dust case,and the aerosol scattering coefficient was higher than the aerosol absorption coefficient.The proportion of PM2.5 rose in the reflux dust case,and the aerosol scattering coefficient corresponded to the aerosol absorption coefficient.

An experiment of 975 nm continuous wave laser irradiating carbon-fiber reinforced resin composites was studied,while the front surface of a target was with and without tangential airflow loadings.The results show that the dynamic responses are obviously different for two different cases.While the front surface of the target is without tangential airflow loading,the ejected inner pyrolysis products ignite surface combustion which is accompanied by dense smoke and flame after laser irradiation for 1.28 s.While the front surface of the target is with tangential airflow loading, there is not obvious combustion flame,but a burn-like bright spot is in the laser-irradiated zone and a little particles spill into the air intermittently.Analysis indicates that the out overflows of pyrolysis products will inhibit the diffusion of oxygen to the target surface,which has a protective effect on the carbon fiber.The presence of tangential airflow not only undermine the protective effect of pyrolysis products on the carbon fiber,but also promote the diffusion of oxygen to the surface.Therefore,the loading of tangential airflow will result in the carbon fiber suffered from oxidative ablation at the low temperature(850 ℃).

As most feature extraction algorithms are highly sensitive to illumination variations,this paper presents a Parameter Adaptive Illumination Robust Extraction Transform(PA-IRFET) based on the histogram of the image.The method includes three steps.Firstly,the gray transformation function and parameters are determined based on the analysis of the histogram of the image.Then,the function is used to transform the image to build an illumination-space of the image.Finally,the illumination-space is combined with feature extraction algorithms to detect illumination robust features.In the experiment of this paper,PA-IRFET is combined with Harris detector to improve the number of detected features of IRFET greatly and the number of repeated features of IRFET has been improved by about 20% on average.This demonstrates that the PA-IRFET is more robust to illumination variations than the IRFET.