基于特征自学习的交通模式识别研究

Transportation mode detection based on self-learning of features

针对目前交通模式识别以人工设计特征为主,特征设计主观性强、区分度不高的问题,本文依据深度学习理论,建立了基于卷积神经网络的特征自动学习模型。该模型利用卷积神经网络自动学习深度特征,然后与人工特征共同用于交通模式识别。模型基于微软Geo Life数据,针对不同特征组合与分类方法设计实验,实验结果表明模型能学习到高区分度深度特征、有效提高交通模式识别准确率。

At present,transportation mode detection is mainly based on artificial design features that have strong subjective design characteristics and low differentiation degree.To solve this problem,a feature self-learning model is established based on convolutional neural network(CNN).Through the CNN′s self-learning of deep features,the model is combined with artificial features before the transportation mode detection is applied.Based on Microsoft′s GeoLife data,the model is designed and tested according to different feature combinations and classification methods.The experiments show that this model can learn deep features with high differentiation degree,thereby effectively improving the accuracy of transportation mode detection.