Research On Cervical Cancer Cell Recognition Technology Based On Cascaded Multi-classifier Fusion

Cervical cancer has always been one of the most common malignant tumors in women,but if treatment is found at an early stage of cervical cancer,the cure may be greatly improved.Many cervical cancer screening techniques are based on manual interpretation of a large number of cell smears by the reading physician,which not only causes great work pressure on the reading physician,but also causes subjective errors in the manual interpretation.At present,the development of automatic cervical cell recognition technology is not mature,and the recognition rate of cervical cancer cells is low.On the basis of previous studies on digital image processing and pattern recognition technology,this paper proposes a cervical cancer cell image classification and recognition method based on cascade multi-classifier fusion,aiming at improving the accuracy of automatic classification and recognition of cervical cancer cell images.Firstly,the cervical cell images were preprocessed.In this paper,according to the characteristics of cell types,the images of cervical cells will be processed by convolution sharpening and histogram equalization,so as to make the edge information of the processed image cell more prominent and improve the classification accuracy.Secondly,the characteristics of preprocessed the cervical cell images were extracted.Since the features extracted by the traditional method contain less useful information,this paper improves the deep learning model by removing the fully connected layer according to the transfer learning idea to extract the feature of the cell image.Since the features extracted from the improved deep learning model contain redundancy,which is not conducive to improving the accuracy of classification,so a new feature extraction model is proposed by combining it with the advantages of the single-variable feature selection model,so that the features extracted from the model can effectively classify cervical cells.Thirdly,the extracted feature data is dimension reduced.In order to improve the accuracy and speed of cervical cell image classification,the linear discriminant analysis(LDA)algorithm was improved in this paper.The Fisher criterion function is improved to make the feature data of the reduced dimension more differentiated.The support vector machine(SVM)classifier is used to further optimize the feature set,and the feature set with large classification effect is retained,which further improves the classification accuracy.Finally,a cascade of multi-classifiers was designed to classify cervical cells.The decision tree classifier,gradient promotion decision tree classifier and Adaboost classifier were constructed respectively.The three classifiers were fused in series using the integration algorithm,and then the three set classifiers were fused into a strong classifier using the weighted voting method to achieve accurate classification and recognition of cervical cell images.In this paper,experiments on pretreatment,feature extraction,feature dimension reduction and classification recognition of cervical cell images were carried out.It can be seen from the experimental results that the proposed method can effectively improve the accuracy of automatic classification and recognition of cervical cancer cell images,and has great application value in the early detection and case diagnosis of cervical cancer cells.