多流卷积神经网络的骨架行为识别

近年来基于骨架数据集的行为识别已成为行为识别领域研究热点.以往基于骨架信息的行为识别研究,是从原始骨架数据中直接提取一系列行为特征信息,作为行为识别网络模型的输入,这种方法忽略了行为中骨架间存在的时空依赖关系,不利于提高行为识别准确度.本文直接利用骨架数据作为网络模型的输入,提出一种基于骨架数据的多流卷积神经网络的行为识别方法.该方法首先进行骨架动作建模;然后利用多流的卷积网络框架提取骨架行为时空特征信息,最后进行行为特征融合和识别.在NTU_RGB+D数据集上进行行为识别,实验结果表明此方法能够提高识别准确率.

Research on the measurement of belt speed by video in coal mine based on improved template matching algorithm

In order to improve the intelligence of video monitoring system of belt and make up the deficiency of higher failure rate and bad real-time performance in the traditional systems of measurement of belt speed, according to the fact that the light of coal mine is uneven, the strength of light changes greatly, the direction of belt movement is constant, and the position of camera was fixed, various algorithms of speed measurement by video were studied, and algorithm for template matching based on sum of absolute differences （SAD） and correlation coefficient was proposed and improved, besides, the tracking of feature regions was realized. Then, a camera calibration method using the invariance of the cross-ratio was adopted and the real-time measurement of belt speed by the hardware platform based on DM642 was realized. Finally, experiment results show that this method not only has advantages of high precision and strong anti-jamming capability but also can real-time reflect the changes of belt speed, so it has a comprehensive applicability.

急性心肌梗死患者采用阶梯式心理护理对其心绞痛发作及负性情绪的影响

目的探究急性心肌梗死患者行阶梯式心理护理对其心绞痛发作和负性情绪的影响。方法择取2018年3月—2019年5月期间该院收治的86例急性心肌梗死患者参与该次研究,通过抽签法进行完全随机分组,包括对照组和观察组,每组43例。对照组患者开展常规护理模式,观察组患者开展阶梯式心理护理干预,对两组患者护理前后分别进行焦虑自评量表(SAS)及抑郁自评量表(SDS)检测,以对患者负性情绪变化进行评定,同时统计比较两组患者心绞痛发作率以及对护理服务的满意度。结果两组患者护理前负性情绪评分经比较差异无统计学意义(tSAS=0.171,tSDS=0.036,P>0.05),护理后,观察组患者负性情绪评分相比对照组明显更低,数据差异有统计学意义(tSAS=15.519,tSDS=17.002,P<0.05);观察组与对照组患者心绞痛发作率分别为2.33%和18.60%,数据差异有统计学意义(χ2=6.081,P<0.05);对护理服务满意度分别为95.35%和76.74%,数据差异有统计学意义(χ2=6.198,P<0.05)。结论阶梯式心理护理能够有效改善急性心肌梗死患者的负性情绪,并降低心绞痛发作风险,还可进一步提升患者对医院护理服务的满意度,有利于构建和谐的护患关系,因此该种护理模式值得在临床中进行大力推广。