Research On Segmentation And Marking Methods Of Daily Activities In Smart Home

In the smart home environment,the deployed sensors are used to perceive the environmental data,and the knowledge of activity recognition is used to process these data,to perceive the needs of residents,to provide help for their lives.Although activity recognition technology has made great progress after long-term research,it still has many challenges in many aspects.For example,in dynamic sensor data flow segmentation,existing technologies sometimes divide the same activity into multiple sensor segments,or combine multiple activities into one activity.Secondly,Secondly,manual marking is not only time-consuming and laborious,but also error prone.In the aspect of feature extraction,only pre-defined features or discrete features are used without paying attention to the time series attributes of sensors.Aiming at the above problems,this paper conducts research in the following three aspects:(1)To solve the problem that it is difficult to determine the window size in the segmentation of sensor data flow,this paper proposes a time-weighted segmentation algorithm based on mutual information of boundary sensors.Firstly,the number of sensors in the training set as the boundary is counted,and then the number of sensors in the whole training set as the boundary for each boundary sensor group is counted,which is used to calculate the proportion of the boundaries of each sensor group.Finally,the time span of each boundary sensor group is calculated.Next,the segmented sensor segment is obtained by segmentation,and then the segmented segment is used for activity recognition.By comparing with the existing methods,the results show that the proposed method has high accuracy and precision.(2)After studying the data marking method,aiming at the problems existing in human marking,this paper improved a semi-supervised data marking algorithm,which firstly used a small number of predefined data to mark a large number of unmarked data through similarity comparison formula,and then carried out clustering marking for unmarked data that did not meet the threshold.By comparing with the existing aspects,it is shown that the data marking algorithm in this paper has a high accuracy in the aspect of active marking.(3)For feature extraction ignore the sensors in time series,this article on the basis of the frequent sequences to join TF-IDF algorithm to optimization activity characteristics,reduce the dimension of feature space,and then select the predefined characteristics of the sensor data and discrete characteristics as feature vectors,recognition and compares them with the existing methods for activity,show that in this paper,the feature extraction algorithm in accuracy,precision rate and recall rate has more advantages.Finally,this paper uses Java to implement the above algorithm,obtains the experimental results from the data set of Weka platform and CASAS,a smart home system of Washington state university,and evaluates and compares the experimental results with relevant evaluation criteria,so as to verify the practicability of this algorithm.