一种基于截断式迁移学习的新冠肺炎识别方法

A Truncated Transfer Learning Based Recognition Method for COVID-19

轻量且高效的肺炎X光(Chest X-Ray,CXR)图像识别模型对于资源受限的平台具有重要意义,为解决以往的研究中很难平衡模型的大小、计算效率和增强性能三者之间的关系的问题,以残差网络ResNet50作为主干网络,并针对医学疾病更加关注中底层抽象特征的特性,采用截断式迁移学习的方法,该方法保留和微调部分的底层,并直接丢弃其他层。同时在截断模块与全连接层中间加入卷积注意力模块,使得模型更加关注病灶区的特征信息,对肺炎图像实现了快速且精确的识别。在重新收集整理的COVID-Xray15k数据集上进行实验,模型分类准确率可达98.6%,与现有研究相比新模型具有更准确且高效的识别新冠肺炎图像性能。

features at the middle and bottom layers.This method preserves and fine tunes some of the lower layers and directly discards other layers.At the same time,a convolutional attention module is added between the truncation module and the fully connected layer to make the model pay more attention to the feature information of the lesion area,achieving fast and accurate recognition of pneumonia images.Experiments are conducted on the newly collected COVID-Xray15k dataset,and the classification accuracy of the model reached 98.6%.Compared with the existing research,the new model has more accurate and efficient performance in recognizing COVID-Xray15k images.