Research On Intelligent Real-Time Risk Early Warning Of Open Pit Blasting Site Based On Improved YOLOv3

As a high-risk,special industry involving explosives,the safety issues during the construction of open pit blasting should not be underestimated.Traditionally,the safety of open pit blasting is entirely the responsibility of the supervisors,however,the complex and changing environment of the open pit blasting site,the complicated preparation work and the many construction links make it difficult for the safety supervision to be fully in place entirely manually.In order to improve mine blasting safety,this paper proposes an intelligent real-time hazard warning method based on improved YOLOv3 for open pit mine blasting sites,which solves the above problems.The main research work of this paper is as follows.(1)The basic situation and safety requirements of the mine are understood through field investigation,and the overall functional framework of the hazard early warning system is designed and related technologies are studied.The present video data compression and transmission technology are studied.Then the target detection algorithm based on shallow learning and the target detection method based on deep learning are analyzed.After the analysis,the image target detection algorithm based on deep learning is used to study the intelligent realtime risk early warning of open-pit mine blasting site.Finally,TensorFlow is chosen to build the deep learning model mentioned in this paper after analyzing and comparing the mainstream deep learning frameworks.(2)The deep learning-based image target detection algorithm is investigated by an experimental comparison method,and a target detection algorithm applicable to the blasting site of open pit mine is derived.Based on the principle analysis of Faster RCNN and Gaussian YOLOv3,the two algorithms are reproduced in Python language;11693 images of daily life scenes and 1660 images of open pit blasting sites are used to compare the training process and detection effect of the two algorithms.Gaussian YOLOv3 performs better in terms of detection accuracy and precision measurement speed,and is more suitable for target detection at open pit blasting sites.(3)Improvements are made to the feature extraction network.the feature extraction network DarkNet-53 used by Gaussian YOLOv3 has the disadvantages of slow detection speed,easy loss of small targets,and easy wrong detection and omission.this paper proposes a singlestage deep neural network PG-YOLOv3,which increases the size of the recorded input image and improves the detection accuracy by a streamlined feature extraction network(Simplified Darknet)to improve the detection accuracy of the model for small targets and increase the detection speed.The target detection model was further optimized by image preprocessing and adjusting the score and IOU later.(4)Realization of real-time video acquisition and transmission at the blasting site of open pit mine.The number of mine blasting projects,short duration,rudimentary site conditions,and frequent blasting site locations make the application of traditional hazard warning systems very difficult.To solve this problem,this paper realizes real-time video acquisition and transmission of blasting sites in open pit mines with hardware such as mobile wireless network cameras,4G wireless routers with H.264 video compression algorithms and RTP real-time transmission protocols.(5)Development of an intelligent real-time hazard warning system for open pit blasting sites.PG-YOLOv3 is used to detect personnel,vehicles and other targets in the monitoring video of open pit mine blasting site in real time,and realize a series of hazard warning functions of open pit mine blasting site according to the detection results,including blasting vehicle speeding warning,hazard detention warning,helmet and mask detection,personnel abnormal behavior alarm,personnel and vehicle statistics,etc.This paper uses C#and Python programming languages,combined with Visual Studio 2015 development platform,to complete the development of the hazard warning system,and finally,analysis and evaluation are conducted.