查询结果:   黄文明,冷金强,邓珍荣,徐双双,雷茜茜.基于反卷积神经网络的脑脊液图像快速识别模型[J].计算机应用与软件,2016,33(7):225 - 228,303.
中文标题
基于反卷积神经网络的脑脊液图像快速识别模型
发表栏目
图像处理与应用
摘要点击数
747
英文标题
FAST CSF IMAGES RECOGNITION MODEL BASED ON DECONVOLUTIONAL NEURAL NETWORK
作 者
黄文明 冷金强 邓珍荣 徐双双 雷茜茜 Huang Wenming Leng Jinqiang Deng Zhenrong Xu Shuangshuang Lei Qianqian
作者单位
桂林电子科技大学计算机科学与工程学院 广西 桂林 541004     
英文单位
School of Computer Science and Engineering,Guilin University of Electronic Technology,Guilin 541004,Guangxi,China     
关键词
反卷积神经网络 特征提取 脑脊液 图像识别 分类线性支持向量机
Keywords
Deconvolution neural network Feature extraction Cerebrospinal fluid Image recognition Classified linear support vector machine
基金项目
广西自然科学基金项目(2013GXNSFAA 019350)
作者资料
黄文明,教授,主研领域:网格计算,图形图像处理,软件工程,信息安全。冷金强,硕士生。邓珍荣,副教授。徐双双,硕士生。雷茜茜,硕士生。 。
文章摘要
提出一种基于反卷积神经网络的脑脊液CSF(Cerebrospinal Fluid)图像快速识别模型。该模型使用无监督学习方法从底层边缘特征到高层对象部分连接对整个图像进行图片特征表述。此外,对不变性、多层模型中层与层如何直接训练等基本问题设计了一系列方法,如引入开关变量,计算每一幅图片试用的滤波器,并允许对每一层的图片单独训练,都提高了学习的鲁棒性。实验结果表明,该模型大大改进了脑脊液细胞图像识别的准确率,同时提高了训练效率。
Abstract
In this paper, we put forward a way that is based on deconvolution neural network for fast recognition of cerebrospinal fluid (CSF) image model. The model uses an unsupervised learning method to represent the features of picture in terms of whole image from the bottom edge features to the connection of high-level object parts. In addition, we design a series of approaches for the basic issues including the invariance and the way of training directly between layers in multi-model, etc., such as introducing the switch variables and computing the trial filters for every image, and allow the single training of the pictures on every layer, these have all improved the robustness of learning. Experimental results show that, the proposed model greatly improves the accuracy of CSF cell images recognition while raises the training efficiency.
下载PDF全文