基于加权深度特征的医学图像并行检索仿真  

Parallel Medical Image Retrieval Simulation Based on Weighted Depth Features

作　　者：毛亚青 王亮 胡俊峰 MAO Ya-qing;WANG Liang;HU Jun-feng(School of Medical Informatics,Xuzhou Medical University,Xuzhou Jiangsu 221000,China)

机构地区：[1]徐州医科大学医学信息学院,江苏徐州221000

出　　处：《计算机仿真》2021年第11期438-444,共7页Computer Simulation

基　　金：国家自然科学基金青年科学基金项目(31900022);教育部产学合作协同育人项目(201801226011);中华医学会医学教育分会、中国高等教育学会医学教育专业委员会2018年医学教育研究立项课题重点项目(2018A-N05065)。

摘　　要：针对传统医学图像检索存储体量大、检索效率低等问题,提出一种基于加权深度特征的医学图像并行检索方法。利用基于部分语义加权的卷积神经网络进行图像深度特征提取,筛选深层网络中对图像信息有效表达的特征;建立图像分布式乘积量化哈希编码模型,实现对高维特征的分布式编码压缩;最后,结合Spark并行模型对查询的医学图像执行分布式并行检索。在医学图像数据集上的实验证明,所提出的方法与传统方法相比具有更好的检索精度和检索效率表现。Aiming at the problems of large storage volume and low retrieval efficiency of traditional medical image retrieval, a parallel medical image retrieval method based on weighted depth features is proposed. The convolutional neural network based on partial semantic weighted aggregation was used to extract the deep features, and the features of effective representation of image information in the deep network were screened. The image distributed product quantization hash coding model was established to realize the distributed coding compression of high-dimensional features. Finally, a distributed parallel search was performed on the medical image of the query in combination with the Spark parallel model. Experiments on medical image datasets show that the proposed method has better retrieval accuracy and retrieval efficiency than traditional methods.

关 键 词：医学图像检索 卷积神经网络 哈希学习 分布式计算 

分 类 号：TP391.3[自动化与计算机技术—计算机应用技术]