多目标注意策略模型及参数影响分析

Multi-objective Attention Strategy Model and Parameter Impact Analysis

文章对注意机制的认知过程进行分析,建立注意决策模型,在此基础上提出3种注意转移策略,即“忽视时长优先转移策略”、“检测难度优先转移策略”和“高等级目标优先转移策略”,并采用Matlab软件对算法进行仿真实验。实验结果表明,若每次均选择被忽视时间最久的目标作为下一个关注焦点,可以获得最短的注意间隔时间,同时目标等待时间也最短。文章同时对影响注意间隔时间的因子进行了讨论,实验证明效率因子&与注意转移速度正相关,调节因子g,与注意转移速度也正相关,将采用“忽视时长优先转移策略”的注意机制与深度学习相融合,可以提高时间序列数据特征识别速度。

In this paper,the cognitive process of the Attention Mechanism is analyzed,the attention decision model is established,and three attention transfer strategies are proposed:the"ignore duration priority transfer strategy",the"detection difficulty priority transfer strategy"and the"high-level goals priority transfer strategy".Meanwhile,the algorithm is simulated with Matlab software.The experimental results show that if the target with the longest neglected time is selected as the next focus each time,the shortest attention interval time and the target waiting time can be obtained.T wo factors affecting the attention interval are also discussed in this paper.It is proved that the efficiency factor and the adjustment factor w;is positively rclated to the speed of attention transferring,furthermore,the combination of the"ignore duration priority transfer strategy"attention mechanism and deep learning can improve the recognition speed of time series data.