融入时序激励机制的人体行为识别

为解决现有人体行为识别方法不能有效融合视频局部特征的问题,提高行为识别的准确率,对视频的一般特点进行分析,提出一种时序激励机制。以BN-Inception作为基础模型分别提取RGB图像序列与光流图像序列的特征,在模型中嵌入时许激励模块实现对视频局部特征序列的动态加权,突出局部特征序列中对行为识别有益的特征的作用,使融合得到的全局特征更具鉴别力。在数据集HMDB51和自建的油田生产现场行为识别数据集OilField-7上进行实验,准确率分别可达71.6%和92.8%,验证了所提方法的有效性。

A Research on Demands Forecasting and Personnel Training of Tourism Talents A Case Study of Zhejiang Province

By analyzing the recent 15 years＇ statistical data of Zhejiang tourism human resources, this paper analyzes the status of Zhejiang tourism talents. ARIMA （Autoregressive Integrated Moving Average） model is a method of time series prediction. This paper predicts the trends of the next three years＇ demands of Zhejiang tourism talents based on ARIMA model in order to promote the tourism in Zhejiang Province. According to the demands forecasting, the number of the employees required by the hotels is 10 times of travel agencies in 2015. At last, some solutions and suggestions are provided such as strengthening the talents training establishing tourism talents mobility mechanism and improving tourism talents excitation mechanism