Research On Object Rapid Detection Method Of Internet Of Thing Terminal Based On Convolutional Neural Network

With the development of the Internet of Things,there are more and more demand for object detection in the application fields of the Internet of Things such as smart transportation and public safety,and the requirements are getting higher and higher.As an important device in the Internet of Things system,the Internet of Things terminal needs to undertake the task of object detection.Convolutional Neural Network(CNN),as a common algorithm in the field of computer vision,plays an important role in object detection scenarios in the field of Internet of Things.However,the convolutional neural network algorithm needs to consume a lot of resources,which makes it difficult to achieve rapid object detection on resource-constrained Internet of Things terminals,and thus cannot meet the actual application requirements of Internet of Things terminals.FPGAs have become a good choice for applying convolutional neural networks to Internet of Things terminals due to their low power consumption,high flexibility,and parallel computing characteristics.To this end,this thesis is based on the Internet of Things terminal of the ARM + FPGA architecture to study the acceleration of convolutional neural networks and the actual application requirements of the Internet of Things terminal.The main research contents are as follows:(1)According to the needs of the Internet of Things terminal to achieve rapid object detection and the function of interacting with the server,an overall design scheme of object detection for the Internet of Things terminal is proposed.(2)In order to achieve rapid object detection on the Internet of Things terminal,a rapid object detection method for the Internet of Things terminal is proposed.First,according to the combined advantages of ARM + FPGA,the hardware and software cooperation strategy of ARM and FPGA is adopted.The hardware acceleration is run on FPGA,and the software logic is run on ARM.Then the acceleration of the convolutional layer in the convolutional neural network is studied,and an acceleration scheme that combines input channel parallelism and output channel parallelism is adopted.(3)According to the convolution parallel strategy combining input channel parallel and output channel parallel,the time model is established.The model takes the storage resources and computing resources of the Internet of Things terminal as constraints,and evaluates the running time of the convolutional layer according to the data transmission time and convolutional calculation time of the convolutional layer in the convolutional neural network.The test results show that the optimal calculation parameters can be obtained by solving the time model according to the terminal resources,thereby reducing the object detection time.In addition to achieving rapid object detection on the Internet of Things terminal,this thesis also uses the interaction between the Internet of Things terminal and the server to update the object detection model parameters on the Internet of Things terminal,so that the Internet of Things terminal can adapt to different Internet of Things application scenarios.The function verification and performance test are carried out on the terminal.The test results showed that the object detection model parameters can be updated on the Internet of Things terminal,and the proposed rapid object detection method can achieve rapid object detection on the Internet of Things terminal.