Cervical cell deep-learning automatic classification method based on fusion features

To overcome the limitations of single-model feature extraction and further improve the accuracy of automatic cervical cell classification, this paper proposes a classification method based on the fusion of features of cervical cells. First, an image set of cervical nuclei after segmentation was constructed and the shallow features of the images were obtained by extracting the shape, chromaticity, and texture features of the nuclei. Then, based on the VGG16 network pretrained by transfer learning, the training image was inputted into the network and the deep features of the image were automatically extracted by the convolution layer. The two types of features were then normalized and spliced to expand the feature dimension and generate new features. Finally, the new feature group was inputted into the classification network for retraining and the classification result after fusion of the features was obtained. The accuracy of the second and seventh classifications could reach 0.981 and 0.923, respectively. The proposed method has practical significance for promoting the automatic application process of cervical cancer screening.