| Population Aging is a global problem at present.Monitoring the fall of the elderly and responding in time is one of the key issues in the intelligent pension field.Distributed optical fiber sensing system can meet the requirements on long-term realtime,environmental complexity and economic benefit for intelligent monitoring of nursing homes or elderly people living alone,and can form beneficial complementarity with intelligent terminal equipment,video monitoring system and other technologies.In particular,the phase sensitive optical time-domain reflectometer（φ-OTDR）distributed optical fiber sensing system has the advantages of long-distance multi-point disturbance location detection,which is suitable for complex environment,real-time monitoring and so on,and has been widely used in the field of disturbance signal measurement and classification early warning.Based on φ-OTDR system and combined with the monitoring scene requirements of elderly people falling down indoors,this thesis aims to solve the common problems of disturbance signal pattern recognition in distributed fiber optical sensing system,such as relatively single acquisition signal and lack of effective method to screen eigenvalues for complex signals,and so on.The specific research contents are as follows:1.The detection status of elderly people fall and other behaviors indoors in intelligent pension field has been investigated,and the strengths and weaknesses of existing technologies have been compared.Also the advantages and disadvantages of various optical fiber sensing systems and their application scope have been discussed.The common problems of optical fiber sensing technology in the field of disturbance signal were clarified.Finally,φ-OTDR system was adopted to study the key problems affecting its fast and accurate pattern recognition and early warning performance.2.Combining the common indoor behavior scenes of the elderly and the lack of complex disturbance signals in term of signal acquisition,2242 groups of signals,including four types,that is walking,falling,walking + falling and environmental sound,were collected finally as basic data for the study.3.Considering the actual needs of indoor positioning in nursing homes and other areas in subsequent application expansion,the collected signals were preprocessed,and the precise positioning within 1 meter is achieved through moving average method and other methods.At the same time,an idea to depict the walking roadmap indoors was proposed,which could provide a pathway to present the real-time movement trajectory.4.Due to the poor recognition effect of low-dimensional eigenmatrix in complex signals,790 dimensions of various eigenvalues were extracted in this paper for subsequence analysis.In view of the lack of effective high-dimensional feature matrix screening algorithms,a three-level feature screening algorithm GA-MI was proposed,which can effectively solve the common problem of feature screening in distributed fiber sensing.5.Different pattern recognition classifiers were used to verify GA-MI screening results,compared with original data sets and screening results of various feature screening algorithms,comprehensively from the accuracy,response time,generality,flexibility and other test criteria.The results showed that the proposed feature screening algorithm can optimize the accuracy and processing time,and has good effect on every different classifiers.It has great practical value as the proposed feature screening algorithm met the actual needs of flexible adjustment of feature dimension.At the same time,the average classification accuracy reached 93.01%,indicating that the scheme proposed in this paper can be applied in the field of intelligent pension field,to solve multiple behavioral pattern recognition problems indoors innovatively.It laid a foundation for application of φ-OTDR distributed optical fiber sensor used in various special occasions of intelligent monitoring. |
PDF Full Text Request