摘要
传统头饰图片识别方法的特征点由研究人员手工提取,工作量大且准确率低,识别系统存在预处理步骤繁琐、样本要求高等缺点。针对上述问题,文中通过构建卷积神经网络从大量图片数据中自动学习头饰图片的深层特征。文中的CNN模型选用稀疏性较好的Re LU激活函数调整输出,利用反向传播算法(BP算法)优化网络参数,在训练得到的CNN模型后接Softmax分类器进行识别。实验结果表明,系统对头饰图片测试集的识别率达到96. 25%,具有良好的识别准确率和识别效率。
The feature point of the traditional headdress image recognition method was extracted by the researchers manually. The system has some disadvantages including tedious preprocessing steps,high sample requirement and low accuracy. To solve these problems,the convolutional neural network was constructed to learn the deep features from image data. The CNN model selected the Re LU function with better sparsity to adjust the output,and used back propagation algorithm to optimize the network parameters. The softmax classifier was identified after the CNN model.The experimental results showed that the recognition rate of the system to the test set of headdress reached 96. 25%.This method was proved to have good recognition accuracy and recognition efficiency.
引文
[1]郑京华.少数民族头饰文化探微[J].中央民族大学学报:哲学社会科学版,1992(2):75-78.Zheng Jinghua.Minority headdress culture exploration micro[J].Journal of Minzu University of China:Philosophical and Social Science Edition,1992(2):75-78.
[2]管彦波.少数民族头饰中的图腾遗迹[J].云南民族大学学报:哲学社会科学版,1995(3):46-48.Guan Y B.Totem relics in minority headdress[J].Journal of Yunnan University of Nationalities:Philosophical and Social Science Edition,1995(3):46-48.
[3]Lee J S,Kuo Y M,Chung P C.The adult image ide-ntification based on online sampling[C].Santiago:International Joint Conference on Neural Networks,IEEE,2006.
[4]Tang X,Stewart W K,Huang H,et al.Automatic plankton image recognition[J].Artificial Intelligence Review,1998,12(1-3):177-199.
[5]Lecun Y,Bengio Y,Hinton G.Deep learning[J].Nature,2015,521(7553):436-447.
[6]Lenz I,Lee H,Saxena A.Deep learning for detecting robotic grasps[J].International Journal of Robotics Research,2013,34(4-5):705-724.
[7]Schmidhuber J.Deep learning in neural networks:an-overview[J].Neural Networks the Official Journal of the International Neural Network Society,2014,61(4):85-117.
[8]Lécun Y,Bottou L,Bengio Y,et al.Gradient based learning applied to document recognition[J].Proceedings of the IEEE,1998,86(11):2278-2324.
[9]Liang M,Hu X.Recurrent convolutional neural network for object recognition[C].Singapore:Computer Vision and Pattern Recognition,IEEE,2015.
[10]Chua L O,Roska T.CNN paradigm[J].IEEE Transactions on Circuits&Systems I Fundamental Theory&Applications,1993,40(3):147-156.
[11]王丽君,于莲芝.基于卷积神经网络的位置识别[J].电子科技,2017,30(1):104-106.Wang Lijun,Yu Lianzhi.Location recognition based on convolution neural network[J].Electronic Science and Technology,2017,30(1):104-106.
[12]Schmidthieber J.Nonparametric regression using deep neural networks with Re LU activation function[J].Arxiv,2017(9):1318-1325.
[13]Mishkin D,Matas J.All you need is a good init[J].Arxiv,2015,69(14):3013-3018.
[14]Li D,Dong Y.Deep learning:methods and applications[M].Boston:Now Publishers Inc.,2014.
[15]高原.基于BP神经网络的文本验证码破解[J].电子科技,2012,25(7):37-42.Gao Yuan.Text verification code cracking based on BP neural network[J].Electronic Science and Technology,2012,25(7):37-42.
[16]Wang Lin,Zeng Yi,Chen Tao.Back propagation neural network with adaptive differential evolution algorithm for time series forecasting[J].Expert Systems with Applications,2015,42(2):855-863.