融合并行投票决策树和半监督学习的鼠标轨迹识别方法

Mouse Trajectory Recognition Method Based on Parallel Voting Decision Tree and Semi-supervised Learning

本文针对已有鼠标轨迹识别方法存在的问题,提出了一种基于并行投票决策树的半监督鼠标轨迹识别方法.首先,本文对鼠标轨迹进行分析,根据多尺度特征思想提取出包括局部轨迹在内的105个特征,并对鼠标轨迹特征进行了划分.其次,本文提出了鼠标轨迹识别的半监督学习方法,避免过拟合和数据噪声的影响.最后,为了提高方法的效率,本文提出并行投票决策树模型,训练多尺度特征,对人的鼠标轨迹和机器鼠标轨迹进行分类.实验结果显示,本文方法具有较好的性能.

Aiming at the problems in the mouse track recognition method，this paper presents a semi-supervised mouse trajectory recognition method based on parallel voting decision tree.Firstly，105 features including global ones，local ones and multi-scale ones are selected and extracted，and then，they are analyzed and divided into groups.Secondly，we propose a semi-supervised mouse tracking recognition method to avoid over-fitting and data noise.Then，to make our method more efficient，we propose a parallel voting decision tree model，and train the multi-scale features using this model.Finally，the experimental results demonstrate that our mouse tracking recognition method works better than some stateofart methods.