您的位置: 首页 > 农业专利 > 详情页

Pattern Recognition System and Mehod of Ultra-Wideband Respiration Data Based on 1-Dimension Convolutional Neural Network
专利权人:
GACHON UNIVERSITY OF INDUSTRY-ACADEMIC COOPERATION FOUNDATION
发明人:
한기태,김성훈,황학인
申请号:
KR1020190074259
公开号:
KR1021341540000B1
申请日:
2019.06.21
申请国别(地区):
KR
年份:
2020
代理人:
摘要:
The present invention relates to a UWB breathing data pattern recognition system based on 1-D CNN, comprising: a pre-processing unit for converting a breathing signal obtained from a UWB radar into a vector; An analysis unit for extracting the transformed vector as a feature vector through 1-D CNN; And a learning/cognition unit that learns and stores and stores the extracted feature vectors through each 1-D CNN, matching each breath pattern, and the analysis unit receiving an input vector of the transformed vector; And a convolutional layer that extracts the feature map by performing convolution of the input vector by 1 * m-type kernel, and serializes the features included in the feature map through Max Pooling, and the learning/cognitive unit uses the serialized features. It includes a Fully-connected Layer that randomly selects (Dropout) to correspond to a preset ratio to learn and recognize features. According to the present invention as described above, by detecting the UWR radar breathing signal, and learning by extracting the feature vector through 1D CNN, any one of general breathing, slow breathing, tachypnea, apnea or movement from the UWR radar breathing signal By modeling the optimal neural network structure and parameters for the pattern features, it is possible to recognize a variety of breathing patterns with high accuracy compared to the conventional one.본 발명은 1-D CNN 기반의 UWB 호흡 데이터 패턴 인식 시스템에 관한 것으로서, UWB 레이다로부터 획득한 호흡 신호를 벡터 형태로 변환하는 전처리부; 변환된 벡터를 1-D CNN을 통해 특징 벡터로 추출하는 분석부; 및 추출한 특징 벡터를 각 1-D CNN을 통해 학습하여 호흡 패턴 별로 매칭하여 저장 및 관리하는 학습/인지부를 포함하되, 분석부는 변환된 벡터를 입력받는 Input Layer; 및 입력된 벡터를 1 * m 형태의 Kernel에 의한 Convolution을 수행하여 Feature map을 추출하고, Max Pooling을 통해 Feature map에 포함된 특징들을 직렬화 하는 Convolutional Layer를 포함하고, 학습/인지부는 직렬화된 특징들을 기 설정된 비율과 대응하도록 무작위로 선택(Dropout)하여 특징을 학습 및 인지하는 Fully-connected Layer를 포함한다.상기와 같은 본 발명에 따르면, UWR 레이다 호흡 신호를 검출하고, 이를 1D CNN을 통해 특징 벡터를 추출하여 학습함으로써, UWR 레이다 호흡 신호로부터 일반호흡, 서호흡, 빈호흡, 무호흡 또는 움직임 중에 어느 하나의 호흡 패턴 특징에 대한 최적의 신경
来源网站:
中国工程科技知识中心
来源网址:
http://www.ckcest.cn/home/
相关发明人
김성훈
한기태
황학인
相关专利

意 见 箱

匿名:登录

个人用户登录

找回密码

第三方账号登录

忘记密码

个人用户注册

必须为有效邮箱
6~16位数字与字母组合
6~16位数字与字母组合
请输入正确的手机号码

信息补充