基于深度卷积神经网络的细胞分类新方法

A New Method of Cell Classification Based on Deep Convolution Neural Network

针对现有的细胞分类方法在准确率方面无法满足人们要求的现象,本文提出一种基于深度卷积神经网络的细胞分类新方法:嵌套残差网络(Multiple Residual Neural Network,M-ResNet).该方法以深度学习理论为基础,在原始ResNet50基础上添加了更高级别的快捷连接(嵌套快捷连接),以挖掘残差网络的优化能力.实验采取宫颈癌细胞作为数据集进行了细胞分类方法测试,其中3528幅作为训练集,350幅作为测试集.通过与ResNet50网络模型进行对比实验,得出测试结果表明:该方法可以有效提高细胞分类的正确率和工作效率,验证了该方法的有效性.这些研究对卷积神经网络的应用和细胞分类方法的发展有着重要的意义,有很好的现实价值.

In view of the fact that the accuracy of existing cell classification methods cannot meet the requirements of people,a new method of cell classification based on deep convolution neural network,Multiple Residual Neural Network,is proposed.Based on the theory of deep learning,a higher level of skip connection(Multiple skip connection)is added to the original ResNet50 to mine the optimization ability of residual network.In the experiment,cervical cancer cells were used as a data set for cell classification,of which 3 528 as training set and 350 as test set.Compared with the ResNet50 network model,the test results show that the method can effectively improve the accuracy and efficiency of cell classification,and verify the effectiveness of the method.These studies are of great significance to the application of convolution neural networks and the development of cell classification methods.