In order to make the machine intelligence reach human's identification capabilities,there is a need to provide a labeled sample which is hard to identify.Efficiency of commonly used methods such as keyboard,mouse,and so on,are low.The labeling method based on eye movement signal is not required to operate manually,but current studies mostly adopt shallow layer models with dependent characteristic projects to achieve interest detection to label samples.Aiming at the problem of shallow-layer model,gaze sequences is classified to achieve interested detection based on feature channel weight redistribution multi-scale residual network model.Effectiveness of this method is verified by comparative experiments.Experimental results show that classification precision of the proposed multi-scale residual network model reaches 96%,compared with existing shallow layer model-based methods and non-modified methods based on deep layer model,the precision and robustness of interested detection are significantly improved.
关键词
视线追踪/眼动事件检测/时间序列分类/多尺度残差网络/感兴趣检测
Key words
gaze tracking/eye movement event detection/time series classification/multi-scale residual network/detection of interest
维普
万方数据