Research And Implementation Of Cervical Cell Image Segmentation Method Based On SOLOv2

In recent years,the prevalence and mortality rate of cervical cancer has become higher and higher,and it has gradually become younger.Early prevention and screening of cervical cancer is of great significance to reduce the incidence.Traditional cervical cancer testing uses cell coating to identify cell types.The whole process relies on doctors to observe the cell morphology to determine whether the disease is sick.There is uncertainty and subjectivity in the entire process.The popularity of artificial intelligence brings new solutions to medical image processing,effectively improving the rate of cell recognition and reducing error rates.However,it also has great requirements for the number of cervical images.Contains noise,resulting in low detection accuracy of the model.Although the existing cervical cell segmentation detection model is faster and more accurate than artificial recognition,it still cannot solve the segmentation of overlapping and blurred cells,and cannot accurately segmentation and identification of overlapping cells.Based on the above problems,this article studies the division of cervical cells and the identification of cell types.The specific research content is as follows:(1)For the scarcity of cervical cell images,the use of Albumentations technology to expand data volume in pixels and space-level aspects,so that deep learning models can fully learn feature information.For the high noise problem contained in the image,this article proposes a algorithm based on double-domain filtering noise,combining median filtering with wavelet transformation,removing noise while retaining complete feature information to reduce image noise interference.(2)In response to the current problem of low cervical cell detection accuracy,this article is based on the SOLOv2 algorithm to improve it.Integrate the ResNeSt network,and add feature enhanced FPT operations after the characteristic pyramid FPN,integrate high-level features with the underlying features,and extract richer feature information.Introduce the deformation convolution,define the value of each sampling point to sieve to remove duplicate information,reduce the calculation amount of subsequent operations,and achieve the effect of improving the model accuracy.Finally,the improvement of the cervical cell detection system is used to use the improved algorithm.