多通道卷积神经网络图像识别方法

发布时间：2018-02-24 14:23

本文关键词： 卷积神经网络多通道梯度图像随机化特征融合分类　出处：《河南科技大学学报(自然科学版)》2017年03期 　论文类型：期刊论文

【摘要】：为了更好地利用图像数据中隐含的特征信息,将多方向梯度信息作为边缘信息的基本表达,提出了一种基于图像梯度的多通道卷积神经网络图像识别方法。先将图像进行Sobel算子处理,得到水平方向、垂直方向及两个对角方向的4个梯度图像。然后,建立4个多层卷积神经网络,学习4个不同方向梯度图像的特征。再将4个不同方向的特征进行随机化特征融合,得到样本的特征后经过批标准化处理。最后,通过分类器得到分类结果。在数据库Cifar-10和MNIST上进行了验证,验证结果表明:本文提出的模型具有较好的泛化能力,相比单通道卷积神经网络,在两个数据库中识别错误率分别降低了9.85%和0.38%。
[Abstract]:In order to make better use of the implicit feature information in the image data, the multi-direction gradient information is regarded as the basic representation of the edge information. A multi-channel convolution neural network image recognition method based on image gradient is proposed. The image is processed by Sobel operator, and four gradient images in horizontal, vertical and two diagonal directions are obtained. Four multilayer convolution neural networks are established to learn the features of four gradient images in different directions. Then the features of four different directions are randomly fused to obtain the features of the samples and then processed by batch standardization. The classification results are obtained by classifier. The results are verified on Cifar-10 and MNIST database. The results show that the proposed model has better generalization ability, compared with single-channel convolution neural network, the proposed model has better generalization ability than single-channel convolution neural network. The recognition error rates in both databases were reduced by 9.85% and 0.38, respectively.
【作者单位】：武汉理工大学自动化学院;
【基金】：国家“863”计划基金项目(2015AA015904)
【分类号】：TP391.41;TP183

【相似文献】