基于深度神经网络的手写数字识别方法研究

Research onhandwritten digit recognition method based on deep neural network

手写数字识别技术的发展随着人工智能的进步也得到了体现,并已渗入到人们的生活中。文章运用深度神经网络模型(DNN),来完成手写数字识别。通过改进代价函数,使用MNIST数据集对模型进行训练,测试评估模型的准确识别率,手动书写阿拉伯数字输入模型进行测试。实验结果表明,数字的正确识别率平均约为98.24%,改进模型在手写数字识别上有较高的准确性,具有一定的使用价值。

The development of handwritten digit recognition technology has been reflected with the advancement of artificial intelligence,and it has increasingly penetrated into people’s lives.The article uses a deep neural network model(DNN)to complete the handwritten digit recognition task.By improving the cost function,the model is trained using the MNIST data set,then the accuracy of the recognition rate of the model is tested,and finally the Arabic numerals are manually written into the model for testing.Experimental results show that the correct recognition rate of digits is about 98.24%on average,and it can be considered that the improved model has good accuracy in handwritten digit recognition.