Detection And Recognition Of Cervical Cancer Cells Based On Multiscale Feature Fusion Convolution Neural Network

With the continuous development of medical image processing and pattern recognition technology,cervical cancer cell detection technology based on digital image processing.This technology is expected to replace the traditional manual screening method,and solve the problems of large workload,high cost,reliability and accuracy influenced by the professional level and subjective emotion of the manual screening method.Since the technology is currently in its infancy,problems such as low equipment intelligence,low recognition accuracy,classification sensitivity and poor specificity are common.In view of the existing problems,this paper has carried out in-depth research on cervical cancer cell detection based on the current research.Firstly,in the automatic acquisition of cervical cell images,the proposed multi-directional gradient accumulative sharpness evaluation algorithm and twostep crawling search algorithm are used to realize autofocus of images.Secondly,unlike the traditional classification method based on accurate segmentation of cytoplasm or nucleus and manual selection operator for feature extraction and reclassification,this paper adopts adaptive threshold algorithm and morphological detection to preprocess the acquired cell image and preprocess it.The latter picture is sent directly to the designed convolutional neural network model for training to complete the classification task.This paper proposes an improved deep convolutional neural network model,which uses a convolutional layer of successively smaller convolution kernels instead of a convolutional layer with a larger convolution kernel,which reduces network parameters and increases the network.Nonlinearity,while adding batch standardization,Dropout and other optimization methods in the network model.Aiming at the problem that the network model can not be effectively studied due to the small number of existing data sets,this paper proposes a small sample data enhancement method.In order to further enhance the classification ability of the network,this paper improves the network model and proposes an improved multi-scale feature fusion convolution neural network model,which combines the feature maps of convolution layers of different sizes,and balances the width of the network while guaranteeing the depth of the network.The experimental results show that the image sharpness algorithm proposed in this paper is better than other traditional algorithms.The improved hill-climbing search algorithm can effectively reduce the search time of the focus point.The proposed multi-scale feature fusion convolutional neural network model based on small sample data enhancement is proposed.The classification accuracy is high and has good sensitivity and specificity.