Research On Liquid Classification By CNN-SVM Model And WIFI Channel Status Information

With the emphasis on public security and the emergence of Intelligence Logistics and Smart Cities in the 21st century,governments,enterprises and even individuals have an more urgent need to know the feature and types of liquids.On the one hand,serval mature technologies have been invented in liquid detection,but all of them are costly which hinders the spread of related products and instruments in our society.On the other hand,more and more related research of channel state information(CSI)based on WIFI was published.These studies even give to commercial WIFI a new function that WIFI has a ability to listen and see as well as the camera equipment,while WIFI not only are far cheaper than camera,but also much easier to obtain in our daily life.This issue has inspired us.Therefore,this paper attempts to use the inexpensive WIFI equipment to complete the task of pattern recognition for liquid detection.This paper attempts to make use of the difference in CSI signal as the judgment type of liquid brought by different WIFI signals through different liquids.Because the CSI data is complex and the CSI of commercial WIFI is easily interfered by the environment and the network card itself,it is relatively difficult to peel off these interferences by calculation or analysis.Therefore,we decided to use self-learning model,CNN-SVM,in CSI classification.In our paper,we proposed three innovations.The one is that we use the channel state information of commercial WIFI to perform pattern recognize the type and feature of liquids.Other is that we use deep learning algorithm for pattern recognition in CSI.Based on analysis and experimentation,we creatively applies CNN-SVM to WIFI liquid detection and based on the channel status information and the characteristics of CNN,this article adopts a distinctive pre-processing method.This paper completed the following work:First,we completed the selection and construction of hardware and software and experimental environment for acquisition and detection equipment.Second,we finished the collection of data sets and collected a total of six different liquids,thousands of samples.Finally,based on the collected data sets,this paper extracts CSI and preprocessed this data.Using the CNN-SVM model for pattern recognition,we found that the accuracy of the CNN-SVM for sample classification is as high as 95%.This accuracy is significant and higher than other algorithms in our experiments.Through the above work,this paper proves that it is possible to use WIFI for liquid pattern recognition.At the same time,it has also proved that the use of CNN-SVM classification in liquid analysis based on WIFI can obtain better results.Our results are pioneering.This article provides a new idea for liquid detection that has been cheaper than before technologies,and has explored a new function for ubiquitous WIFI.Although it is far from enough to practical application only by this paper's conclusion,our conclusion is helpful for feature research and application.