基于多级金字塔卷积神经网络(MLPCNN)的快速特征表示方法

近年来,在机器视觉中基于卷积神经网络(CNN)的特征提取方法取得了令人惊叹的成果,主要原因是深度学习在多层和低维的特征表示上有着很大的优势。但是由于在大尺度图像中卷积滤波的过程速度过慢,导致CNN参数调节困难、训练时间过长,针对这一问题,本文基于传统卷积神经网络(TCNN, Traditional convolution neural network)提出一种快速有效的多级金字塔卷积神经网络MLPCNN(Multi-level pyramid CNN)。这一网络使用权值共享的方法将低级的滤波权值共享到高级,保证CNN的训练只在较小尺寸的图像块上进行,加快训练速度。实验表明,在特征维数比较低的情况下,MLPCNN提取到的特征比传统的特征提取方法更加有效,在Caltech101数据库上,MLPCNN识别率达到81.32%,而且训练速度较TCNN网络提高了约2.5倍。

A Multi-level Pyramid Convolution Neural Network based Fast Feature Representation Method

In recent years, based on convolution neural network (CNN) feature extraction, machine vision has achieved amazing results. Mainly due to deep learning has a great advantage in multi-layers and low-dimensional feature representation. However, because of the speed of convolution filter in processing large-scale image too slow, the parameter adjustment is much difficulty and the training time is too long. To solve this problem, based on traditional convolution neural network (TCNN), this paper proposed a fast and efficient multi-level pyramid convolutional neural network (MLPCNN). This network shared the filter weights of lower to higher using weights-shared methods to ensure that the training of CNN only occur on small size image patches. This would speed up the training of CNN network. Experiments show that, in the case of the feature dimension is relatively low, the feature extraction method of MLPCNN is more effective than traditional feature extraction methods. On Caltech101 database, MLPCNN is capable of achieving high recognition accuracy of 81.32%, and the training time is 2.5 times faster than TCNN network.