Design And Implementation Of Internet On Things Fusion Gateway

With the maturity and development of various conmmunication technologies and the sinking of communication costs,the Internet of Things industry has developed rapidly in recent years.In 2017,the global device access quantity reached 8.4 billion,surpassing the population for the first time.Since 2017,the Internet of Things has been applied in various fields including transportation,environmental protection,logistics,electric power,medical care,security and so on.According to Gartner report,the total number of IoT device access will reach 20 billion by 2020.The explosive growth of access devices and the increasing demand for that have brought new challenges to data analysis and data processing,such as multiple protocol access,the rapid growth of network load,the exponential rise of server-side computing,and the growing of power consumption.This thesis will focus on the first two issues,one is aiming at designing a gateway system completing the fusion of edge nodes of different protocols,one is integrating edge computing into the gateway within limited resources.The gateway system includes sensing front-end hardware adaptation layer,such as temperature,humidity,distance,and other types of signal messages,as well as image information such as images and voices,and various protocol access layers such as WIFI,LoRa,ZigBee,message processing layer,log layer,CNN network acceleration layer,MQTT message publish/subscrib layer,MQTT layer.Based on this gateway system,this thesis will focus on the analysis,quantization optimization of neural networks and model transformation.As for the main contribution,this thesis completed the model convertion design based on Tensorflow neural network model and designed high performance message payload encode and decode.One of the case is face detection under limited resources,and the detection result is then reported to the server to complete the extraction of video structured frame data.particularly,this thesis proposed a piecewise linear quantization method to realize the quantitative optimization of the model,and gived the optimization algorithm and test results.