摘要
为了有效提高宫颈癌的诊断准确率,提出一种基于SE-Res2Net网络的宫颈癌超声肿瘤特征提取技术。在YOLOv3算法模型的基础上,将SE模块嵌入Res2Net网络中,创建一种能够替换原特征提取网络的SE-Res2Net网络,使模型的特征提取能力得到提升。利用重新构建的下采样模块,保证了下采样操作后信息的完整性。将密集连接网络与残差连接网络相结合,组建Res-DenseNet网络以改进YOLOv3模型的原有残差连接方式。实验结果表明,该方法的性能明显优于传统YOLOv3算法,适于在临床诊断中普及应用。
In order to effectively improve the diagnostic accuracy of cervical cancer,a feature extraction technology of cervical cancer ultrasound based on SE-Res2Net network is proposed.Based on the model of YOLOv3 algorithm,the SE module is embedded in the Res2Net network to create a SE-Res2Net network which can replace the original feature extraction network,so that the feature extraction ability of the model is improved.Besides,The reconstructed downsampling module ensures the integrity of the information after the downsampling operation.In order to improve the original residual connection mode of YOLOv3 model,the Res-DenseNet network is constructed by combining the dense connection network with residual connection network.Experiment results show that the performance of the proposed method is significantly better than that of traditional YOLOv3 algorithm,and it is suitable for clinical diagnosis.
作者
张海艳
李洁
张博学
刘静
唐雪蕊
ZHANG Hai-yan;LI Jie;ZHANG Bo-xue;LIU Jing;TANG Xue-rui(Tangshan Workers’Hospital,Tangshan 063000,Hebei Province,China)
出处
《信息技术》
2022年第5期177-182,共6页
Information Technology