基于视频深度学习的小鼠恐惧情绪识别与分析方法研究

Research on Mouse Fear Recognition and Analysis Method Based on Video Deep Learning

在小鼠情绪分析实验中,实验者通常采用机器学习的方法预测小鼠关键部位的位置变化来判定小鼠是否处于恐惧情绪行为。为了进一步提高小鼠恐惧情绪识别和分析的准确率,本文构建了一种高效的轻量化U-Net语义分割模型,提出了根据视频中帧画面之间小鼠轮廓区域的变化幅度来量化分析小鼠恐惧情绪行为的方法。经实验验证,本文方法与专家统计结果的皮尔森相关性系数达到86%以上,证明了本文方法对于小鼠恐惧情绪行为的分析具有较高的准确率,同时对其它小型模式动物行为分析也有一定的参考价值。

To determine whether the mouse is in fear emotion behavior studies,deep learning method is often used to obtain the displacements of key parts of mouse body from frames.In order to improve the accuracy of mouse fear emotion recognition,we established an efficient lightweight U-Net model for semantic segmentation to quantify mouse’s fear emotion behavior based on the variation of mouse contour from frames.With experimental validation,Pearson correlation coefficient between the results from the proposed method and the ground-truth by experts is up to 86%,so that the accuracy of our method is acceptable for mouse fear emotion analysis.In addition,this method offers potential applications for behavior analysis of other model animals.