基于最小类内方差的伪三维残差网络

作为一种提取视频时空特征的深度学习方法,伪三维残差网络(pseudo-3D residual net,P3D ResNet)利用SVM目标函数来驱动深度网络学习,这样该方法继承了SVM的不足——仅考虑了不同类别间的间隔,忽略了同类样本数据的分布信息.针对该问题,提出了基于最小类内方差的伪三维残差网络方法,不仅体现了大间隔原理,同时又利用了样本数据的分布信息.该方法首先使用P3D ResNet提取的特征向量计算类内散度矩阵;然后利用该矩阵构建了新的目标函数;最后通过新构建的目标函数来驱动P3D ResNet的学习.将该方法应用到行为识别领域,多个数据集上的实验结果表明,相比于传统的P3D ResNet,所提出的方法获得了更高的识别准确率,体现出了更好的泛化性能.

Pseudo-3D residual network based on minimum intra-class variance

As a deep learning method for extracting video spatio-temporal features, pseudo-3D residual net(P3D ResNet) uses the objective function of SVM to drive the learning of deep network. In this way, this method inherits the insufficiency of SVM, which only considers the interval between different categories, and ignores the distribution information of similar samples. Aiming at this problem, this paper proposed an improved method called P3D ResNet based on minimum intra-class variance. This method not only embodied the principle of large interval, but also used the distribution information of sample data. First, the method used the feature vector extracted by P3D ResNet to calculated the intra-class divergence matrix. Then it used the matrix to construct a new objective function. Finally, it drove the learning of P3D ResNet by the newly constructed objective function. This paper applied the method to the field of behavior recognition. Experimental results on multiple datasets show that compared with the traditional P3D ResNet, the proposed method achieves higher recognition accuracy and shows better generalization performance.