Design And Implementation Of Safety Helmet Intelligent Monitoring System Based On Deep Learning

In recent years,the rapid development of technologies such as big data analysis,cloud computing,edge computing and the Internet has laid a solid foundation for the broad application of artificial intelligence.Deep learning provides a very reliable solution for target detection because of its great potential in intelligence.Deep learning based on convolutional neural networks is developing rapidly in target detection and is gaining widespread use and research in the industry.However,deep learning has a high threshold for researchers in non-computer-related fields,making it difficult to use.There is a particular efficiency problem for researchers to deploy the model from model training to one click.Therefore,to address these problems,this thesis designs and implements an intelligent detection system based on Io T,cloud computing,edge computing,and deep learning,which includes a cloud platform and an edge platform.The system aims to help users provide fast training and deployment of deep learning.First,the helmet dataset images have been labelled position and classes in this thesis,and then based on different YOLOv5 model algorithms,a model training experiment was carried out on the helmet dataset,and analyzed the detection accuracy and efficiency of the algorithm.After the model training analysis was completed,the model was converted and quantized.Finally,the experimental analysis was carried out on Firefly AIO-3399-Pro C and Firefly Core-1126-JD4 edge devices,and the inference time of different models was obtained.The time difference between model quantization and non-quantization was more than 4 times.Considering the model detection accuracy and the edge Due to limitations such as limited computing resources of the device,this thesis has determined the model of YOLOv5s-ghost as the algorithm of the subsequent intelligent detection system.Then,this thesis analyzed the requirements and functions of the intelligent monitoring system and designed and implemented the functions and pages of the system.The system is built around the life cycle of deep learning.The system is divided into two parts,the edge platform and the cloud platform.The prediction function of the model is realized on the edge platform and the prediction results are uploaded to the cloud platform; the data set production and algorithm are completed on the cloud platform.Main functions such as design and model training evaluation.For security reasons,edge devices need to be registered on the cloud platform before they can issue models.Finally,this thesis tested the implemented intelligent monitoring system.After tested the functions and performance of the cloud platform and the edge platform,the communication between the edge platform and the cloud platform was tested.After the complete test,it was proved that the system has good functions and can fulfil the requirements,and the system has good usability.