基于深度神经网络的手写数字识别模拟研究
|
摘要
当前图像识别大多采用基于特征提取的传统机器学习方法与卷积神经网络的方法,但传统图像识别技术需要手动提取图片特征,而卷积神经网络对硬件要求高,训练时间长等。针对以上问题,提出基于深度神经网络模型的手写体图像识别方法,让机器自动学习特征,并在此基础上,通过改进成本函数,加入dropout防止过拟合,来提高手写数字识别的识别率。仿真实验对比结果表明,基于深度神经网络模型的方法比当前传统算法的识别率提高了3. 41%,有效解决了人工识别费力耗时问题,对手写数字的研究具有重要意义。
The traditional machine learning method based on feature extraction and the convolutional neural network method are mostly used in image recognition at present. However,the traditional image recognition technology requires manual extraction of image features,while convolutional neural network has high requirements on hardware and long training time. In view of the above problems,a handwritten image recognition method was proposed based on the deep neural network model,which enables the machine to learn the features automatically. On this basis,it improves the recognition rate of handwritten numeral recognition by improving the cost function and adding dropout to prevent overfitting. The results of simulation experiment show that the method based on the deep neural network model has improved the recognition rate by 3. 41% compared with the current traditional algorithm,which effectively solves the problem of labor and time consuming in manual recognition.
The traditional machine learning method based on feature extraction and the convolutional neural network method are mostly used in image recognition at present. However,the traditional image recognition technology requires manual extraction of image features,while convolutional neural network has high requirements on hardware and long training time. In view of the above problems,a handwritten image recognition method was proposed based on the deep neural network model,which enables the machine to learn the features automatically. On this basis,it improves the recognition rate of handwritten numeral recognition by improving the cost function and adding dropout to prevent overfitting. The results of simulation experiment show that the method based on the deep neural network model has improved the recognition rate by 3. 41% compared with the current traditional algorithm,which effectively solves the problem of labor and time consuming in manual recognition.
引文
1 Waris M A,Iosifidis A,Gabbouj M. Object proposals using CNNbased edge filtering[C]//International Conference on Pattern Recognition. New York:IEEE,2017:627-632
2 Zhang P,Zhuo T,Huang W,et al. Online object tracking based on CNN with spatial-temporal saliency guided sampling[J]. Neurocomputing,2017,257:115-127
3 Smirnov E A,Timoshenko D M,Andrianov S N. Comparison of regularization methods for image net classification with deep convolutional neural networks[J]. Aasri Procedia,2014,6(1):89-94
4 韩晓艳,赵东.基于粒子群的支持向量机图像识别[J].液晶与显示,2017,32(1):69-75Han Xiaoyan,Zhao Dong. Image recognition based on particle swarm support vector machine[J]. Liquid Crystal and Display,2017,32(1):69-75
5 Kumar M,Jindal M K,Sharma R K,et al. Offline handwritten numeral recognition using combination of different feature extraction techniques[J]. National Academy Science Letters,2018,41(1):29-33
6 Prasad B K,Sanyal G. Multiple hidden markov model post processed with support vector machine to recognize english handwritten numerals[J]. International Journal on Artificial Intelligence Tools,2018,27(5):1-19
7 张颖,戚世蕾,方有亮.基于支持向量机的钢框架结构损伤定位试验研究[J].科学技术与工程,2018,18(7):25-30Zhang Ying,Qi Shilei,Fang Youliang. Experimental study on damage location of steel frame structure based on support vector machine[J]. Science Technology and Engineering,2018,18(7):25-30
8 张雪芹,方婷,李志前,等.基于类Haar特征和Ada Boost的车辆识别技术[J].华东理工大学学报(自然科学版),2016,42(2):260-265Zhang Xueqin,Fang Ting,Li Zhiqian,et al. Vehicle identification technology based on similar Haar features and AdaBoost[J]. Journal of East China university of Science and Technology(Natural Science Edition),2016,42(2):260-265
9 段萌,王功鹏,牛常勇.基于卷积神经网络的小样本图像识别方法[J].计算机工程与设计,2018,39(1):224-229Duan Meng,Wang Gongpeng,Niu Changyong. A small sample image recognition method based on convolution neural network[J].Computer Engineering and Design,2018,39(1):224-229
10 常亮,邓小明,周明全,等.图像理解中的卷积神经网络[J].自动化学报,2016,42(9):1300-1312Chang Liang,Deng Xiaoming,Zhou Mingquan,et al. Convolution neural network in image understanding[J]. Journal of Automation,2016,42(9):1300-1312
11 Zhang X Y,Yin F,Zhang Y M,et al. Drawing and recognizing Chinese characters with recurrent neural network[J]. IEEE Transactions on Pattern Analysis&Machine Intelligence,2018,99:849-862
12 傅天驹,郑嫦娥,田野,等.复杂背景下基于深度卷积神经网络的森林火灾识别[J].计算机与现代化,2016(3):52-57Fu Tianju,Zheng Change,Tian Ye,et al. Identification of forest fire based on deep convolutional neural network in complex background[J]. Computer and Modernization,2016(3):52-57
2 Zhang P,Zhuo T,Huang W,et al. Online object tracking based on CNN with spatial-temporal saliency guided sampling[J]. Neurocomputing,2017,257:115-127
3 Smirnov E A,Timoshenko D M,Andrianov S N. Comparison of regularization methods for image net classification with deep convolutional neural networks[J]. Aasri Procedia,2014,6(1):89-94
4 韩晓艳,赵东.基于粒子群的支持向量机图像识别[J].液晶与显示,2017,32(1):69-75Han Xiaoyan,Zhao Dong. Image recognition based on particle swarm support vector machine[J]. Liquid Crystal and Display,2017,32(1):69-75
5 Kumar M,Jindal M K,Sharma R K,et al. Offline handwritten numeral recognition using combination of different feature extraction techniques[J]. National Academy Science Letters,2018,41(1):29-33
6 Prasad B K,Sanyal G. Multiple hidden markov model post processed with support vector machine to recognize english handwritten numerals[J]. International Journal on Artificial Intelligence Tools,2018,27(5):1-19
7 张颖,戚世蕾,方有亮.基于支持向量机的钢框架结构损伤定位试验研究[J].科学技术与工程,2018,18(7):25-30Zhang Ying,Qi Shilei,Fang Youliang. Experimental study on damage location of steel frame structure based on support vector machine[J]. Science Technology and Engineering,2018,18(7):25-30
8 张雪芹,方婷,李志前,等.基于类Haar特征和Ada Boost的车辆识别技术[J].华东理工大学学报(自然科学版),2016,42(2):260-265Zhang Xueqin,Fang Ting,Li Zhiqian,et al. Vehicle identification technology based on similar Haar features and AdaBoost[J]. Journal of East China university of Science and Technology(Natural Science Edition),2016,42(2):260-265
9 段萌,王功鹏,牛常勇.基于卷积神经网络的小样本图像识别方法[J].计算机工程与设计,2018,39(1):224-229Duan Meng,Wang Gongpeng,Niu Changyong. A small sample image recognition method based on convolution neural network[J].Computer Engineering and Design,2018,39(1):224-229
10 常亮,邓小明,周明全,等.图像理解中的卷积神经网络[J].自动化学报,2016,42(9):1300-1312Chang Liang,Deng Xiaoming,Zhou Mingquan,et al. Convolution neural network in image understanding[J]. Journal of Automation,2016,42(9):1300-1312
11 Zhang X Y,Yin F,Zhang Y M,et al. Drawing and recognizing Chinese characters with recurrent neural network[J]. IEEE Transactions on Pattern Analysis&Machine Intelligence,2018,99:849-862
12 傅天驹,郑嫦娥,田野,等.复杂背景下基于深度卷积神经网络的森林火灾识别[J].计算机与现代化,2016(3):52-57Fu Tianju,Zheng Change,Tian Ye,et al. Identification of forest fire based on deep convolutional neural network in complex background[J]. Computer and Modernization,2016(3):52-57