Research On Image Recognition Of Cervical Cancer Cells Based On Deep Learning

Deep learning is an indispensable part in the development of artificial intelligence,and is widely used in speech recognition,image recognition,face recognition and other fields.Deep learning has turn into a study focus in the area of image recognition.Currently,the majority of the classification and recognition of cervical cancer cell images be part of the shallow learning.Due to the lack of ability of shallow learning to fit complex functions,besides the complexity and difference between cervical cancer cells and the shape of tissues and organs are irregular,moreover the image data of cervical cancer cells have high-order statistical characteristics,leading to a large amount of redundant information between cervical cancer cells.Therefore,this paper deeply studies the convolutional neural network,and improves the algorithm.Moreover,the improved algorithm based on convolutional neural network is applied to the recognition of cervical cancer cell images.The accuracy of image recognition depends on the accuracy of feature extraction.Therefore,feature extraction of images and correct classification are the main content of image recognition research.In this paper,an improved algorithm based on convolutional neural network is proposed.Firstly,a convolution neural network is constructed on the basis of VGG16,and in the stage of down sampling an improved pooling model is proposed.In the down sampling process,each element in the pooling area is assigned appropriate weights to obtain a down sampling feature map,which effectively avoids the loss of key features caused by the max pooling method and the weakening of large eigenvalues caused by average pooling.So that different pooling areas can extract more accurate features.And the improved pooling algorithm is applied to the recognition of cervical cancer cell images.The experimental results indicate that the improved method based on the convolutional neural network reduces the recognition error rate of cervical cancer cell images.The recognition accuracy of cervical cancer cell images can reach 95.61%.In this paper a framework of residual neural network is built,and the gradient descent algorithm is analyzed and improved.An improved algorithm based on residual neural network is proposed.By adding momentum parameters to stochastic gradient descent algorithm,the parameter updates does not completely depend on the previous gradient,so as to reducing the impact of noise.Each time the parameters are updates,several groups of samples are randomly selected for iteration to prevent the solution from falling into local optimum.The improved algorithm is applied to the recognition of cervical cancer cell image.The experimental results indicate that the improved method based on the residual neural network has a high recognition rate,which reduces the recognition error rate of cervical cancer cell images.The recognition accuracy of cervical cancer cell images can reach 96.83%.In summary,the two improved algorithms proposed in this paper have improved the recognition rate of cervical cancer cell image in varying degrees.Therefore,the application of deep learning method in cervical cancer cell image recognition has broad application prospects.