Research Of Single Cell Image Classification Based On Same Layer Multi Scale Kernel Convolution Neural Network

Using machine learning to image classification is one of the current hot research direction.Human was beaten by Google ’s artificial intelligence program Alpha Go stimulates the further development of machine learning.Convolutional neural network is widely used machine learning algorithms,now it has wide application in many fields,such as financial risk early warning,decision-making,character recognition,image classification.Medical imaging technology with the progress of science and technology has been widely applied in clinical practice,so many diseases can be found earlier and earlier treatment,medical image classification mainly depends on the human eye,time-consuming,and is easily affected by the subjective factors of doctors.Using computer for medical cell image classification has become a popular and difficult research.In this paper machine learning algorithm which based on the convolution neural network model is studied,and applied to single cell image classification,The mainly contents are as follows,1,In this paper,the theoretical basis of convolutional neural network is presented,and the process of the evolution from the perceptron model to the convolutional neural network is introduced,and the theoretical knowledge of the Softmax classifier is provided,which provide a theoretical basis for the establishment and improvement of model.2,Due to the small number of images of the open HEp-2 cell image and the cervical cell image datasets,the neural network model can not be trained directly.In this paper,the method of seam carving and high order interpolation is used to scale the image,so single cell image has the same size.At the same time,the method such as cutting,rotation,changing contrast and brightness,standardized is applied on the cell image to enhancement the data sets.3,Aiming at the problem of classical convolutional neural network can not be used for single cell image classification effectively.An improved convolutional neural network model is designed,and the convolution neural network model is improved through six aspects:batch technology,such as local response normalization,Softmax and so on.4.In order to improve the ability of classification and recognition of single cell image based on convolutional neural network model,considering that the receptive field is not fixed in the human visual process,An improved same layer multi scale kernel based convolution neural network model is designed,which can be further improved by increasing the multi-scale receptive field,using ReLUs,changing the number of kernel functions and so on.5,Using Google Tensorflow framework to program the model,and through the ICPR2012 HEp-2 data sets and cervical cell set classification simulation test.The experimental results show that the two models are designed in this paper has good robustness and immunity for incomplete or different contrast and brightness or rotation image,they can well complete the six classification of HEp-2 single cell image and two classification of cervical cell image.