基于CNN-GRNN模型的图像识别

发布时间：2018-03-07 03:38

本文选题：卷积神经网络　切入点：广义回归神经网络　出处：《计算机工程》2017年04期 　论文类型：期刊论文

【摘要】：卷积神经网络(CNN)模型在图像识别中取得了良好的效果,但其识别精度还有进一步提升的空间。为此,设计一种新的图像识别模型CNN-GRNN。利用CNN提取样本图像中的多层次特征信息,将广义回归神经网络代替反向传播神经网络,以提高分类器的泛化能力和鲁棒性,通过均方差和降梯度法训练模型。基于COIL-100和手势库的实验结果表明,与灰度共生矩阵、HU距方法、CNN和CNN-SVM模型相比,CNN-GRNN的识别率分别提升了42.2%,13.43%,3.99%和1.86%,并具有较好的实时性。
[Abstract]:The convolutional neural network (CNN) model has achieved good results in image recognition, but its recognition accuracy still has room for further improvement. A new image recognition model, CNN-GRNN, is designed to extract multi-level feature information from the sample image by using CNN. In order to improve the generalization ability and robustness of classifier, generalized regression neural network is replaced by backpropagation neural network. The model is trained by the method of mean-variance and descending gradient. The experimental results based on COIL-100 and gesture database show that, The recognition rate of CNN-GRNN is improved by 42.2%, 13.433.99% and 1.86%, respectively, compared with the gray co-occurrence matrix HU distance method and CNN-SVM model, and it has better real-time performance.
【作者单位】：昆明理工大学信息工程与自动化学院;
【基金】：国家自然科学基金(61263017) 云南省人才培养项目(KKSY201303120) 云南省教育厅科学研究基金(2014Y086)
【分类号】：TP391.41

【相似文献】