融合图嵌入的Smooth-PCANet图像识别算法

主成分分析网络（Principal Component Analysis Network，PCANet）是一种简单的深度学习算法，在图像识别领域具有优秀的性能。然而PCANet在构建网络卷积核时只关注了图像的主分量信息，忽视了近邻像素点之间的位置关系。而通常情况下，图像的相邻像素点具有空间强相关性，因此利用图结构保持像素点的位置近邻关系更有利于网络提取有效特征。因此，我们将图嵌入思想融入PCANet，提出一种新的图像识别算法Smooth-PCANet。为了验证Smooth-PCANet算法的有效性，我们在人脸、手写体字符以及图片等不同数据集上构建实验，并将Smooth-PCANet与多种基于深度学习的图像识别算法作了对比。实验结果证明，Smooth-PCANet算法比PCANet获得了更高的识别性能，并且更有效地避免了过拟合，在小样本训练时具有显著优势。

Graph Embedding Based Smooth-PCANet for Image Recognition

Principal component analysis network (PCANet) is a simple deep learning algorithm with excellent performance in the field of image recognition. However, when constructing the convolution kernel, PCANet only pays attention to the principal components of the image and ignores the position relationship between the neighboring pixels. In general, the neighboring pixels of the image have strong spatial correlation, so it is more conducive to extract effective features by using the graph to maintain the position information of the pixels. Therefore, we integrate graph embedding into PCANet and propose a new image recognition algorithm, Smooth-PCANet. In order to verify the effectiveness of the Smooth-PCANet algorithm, adequate experiments are performed on different data sets such as face, handwritten characters, and images. Compared with several image recognition algorithms based on deep learning, the experiments demonstrate that the Smooth-PCANet performs better than the PCANet, and avoids overfitting more effectively in small sample training.