基于Keras卷积神经的肺炎识别研究应用  

Research on Pneumonia Identification Based on Convolutional Neural Network with Keras

作　　者：孙自梅 SUN Zimei(College of Computer and Communication Engineering,Nanjing Tech University Pujiang Institute,Nanjing 210000,China)

机构地区：[1]南京工业大学浦江学院,计算机与通信工程学院,江苏南京210000

出　　处：《微型电脑应用》2024年第10期227-231,共5页Microcomputer Applications

摘　　要：以TensorFlow为框架,以Keras为高阶应用程序接口,使用卷积神经网络作为训练模型,设计一套快速有效针对病毒性肺炎的识别系统。主要采用卷积神经网络模拟人的大脑不断学习辨别的过程,包括对图片的预处理、特征提取、数据归一化、模型搭建、TensorBoard集验证、图形化展示等。肺炎识别系统通过采集肺部CT图像,经过图片预处理后进入卷积神经网络训练,再经过验证集的验证,识别病毒性肺炎,其准确率高达90%。为了进一步提高检测精准度,通过调整Dropout参数,使病毒性肺炎识别的准确率提升到了98%左右,此改进方法取得了较大的进步。This paper uses TensorFlow as the framework and Keras as the high-level application programming interface and em-ploys convolutional neural network as the training model to design a fast and effective identification system for viral pneumonia.The design primarily uses convolutional neural network to simulate the continuous learning and discrimination process of the human brain,including preprocessing of images,feature extraction,data normalization,model construction,TensorBoard set validation,and graphical display.The pneumonia identification system collects lung CT images,which are fed into the convolu-tional neural network for training after preprocessing.The validation of the validation set leads to the identification of viral pneumonia with an accuracy rate of up to 90%.To further improve detection precision,this paper adjusts the Dropout parame-ter,improves the accuracy of viral pneumonia identification up to about 98%,marking a significant improvement in this method.

关 键 词：Keras 卷积神经网络 训练模型 肺部CT图像 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]