区域渐近增强与感受野结合的多核近似学习网

为充分提取图像中可辨识信息、提高分类正确率,提出多核近似学习网,该网络主要由2部分构成。在特征提取部分,利用二维高斯分布对原始图像进行区域渐进增强,局部感受野和全局感受野被用于充分提取原始图像和区域渐进增强图像中的局部和全局特征,并将其串联以组成代表图像的特征向量。在分类部分,提出多核近似算法,将近似核映射编码出的低秩特征矩阵作为网络的隐藏层,以求解网络的输出权重。为验证该网络的有效性,利用USPS、MNIST和NORB数据集进行实验,实验证明所提出的多核近似学习网能够在局部感受野极端学习机的基础上进一步提取出特征信息,有效提高了分类正确率。

Multiple Kernel Empirical Learning Network with Gradually Enhanced Region and Receptive Field

In order to fully extract the identifiable information in the image and improve classification accuracy rate, a structure named multiple kernel empirical learning network(MKELN) was proposed. In the feature extraction part, the original image is gradually enhanced by the two-dimensional Gaussian distribution. The local receptive field and the global receptive field are used to fully extract features in the original image and the gradually regional enhancement image, and they are connected in series to form feature vector that represents an image. In the classification part, a multiple kernel empirical algorithm was proposed, and the low rank feature matrix is used as the hidden layer of the network to solve the output weight of the network. To verify the effectiveness of this network, it was tested with USPS, MNIST and NORB data sets. The experiment proves that the proposed MKELN can further extract feature information based on ELM-LRF, effectively improving the classification accuracy.