Research On Unsafe Behavior Detection And Safety State Analysis Of Construction Site Based On UAV Remote Sensing Data

Most water conservancy and hydropower projects are built in complex engineering geological environment,with obvious regional characteristics such as high,slippery and steep.The inspection of unsafe behavior of on-site personnel is faced with problems such as low work efficiency,high labor cost and many safety risks.The planning and utilization of engineering environmental conditions such as construction sites and on-site roads,normalized safety inspection and safety status analysis also need to be dynamically adjusted in time in combination with engineering progress.Unmanned Air Vehicle(UAV)has the advantages of wide field of vision,work efficiency not restricted by terrain factors,and strong data acquisition ability.The use of drones can quickly obtain aerial images and remote sensing data,providing data support for unsafe behavior detection and safety status analysis at the construction site.Based on the application background of Zhushou Reservoir Expansion Project,this paper discusses the application of UAV remote sensing data to the on-site unsafe behavior inspection and road safety status analysis of water conservancy and hydropower projects.By introducing the target detection technology,the target detection model is constructed,and the UAV technology is combined to improve the efficiency of the project site inspection.Based on the UAV remote sensing data,the numerical calculation model of road rock and soil stability analysis is constructed.With the help of numerical simulation method,the safety state of road rock and soil is analyzed to ensure the safety of project construction.The main research work and achievements are as follows:(1)UAV remote sensing data acquisition based on hierarchical encryption measurement method.Combined with the terrain conditions of Zhushou Reservoir dam site,a UAV data acquisition method for layered encryption measurement is designed.By setting the safe flight height of the UAV in a specific scene and changing the flight height in the area where the obstacle exists,the UAV data is encrypted and collected,and the UAV remote sensing data meeting the accuracy requirements is obtained.(2)Construction of UAV target detection data set for unsafe behavior recognition and target detection at construction site.Using the image annotation tool Label Img,the human and vehicle targets in the obtained UAV remote sensing data are labeled,and the computer-learnable information of the target object is obtained.The image cropping program is compiled and implemented based on Python language for data augmentation.The cropped data is processed according to the COCO(Common Objects in Context)data set format to make a standard COCO format data set.(3)The YOLOv5 baseline network is lightweight improved from the aspects of feature extraction network,spatial pyramid pooling structure and feature fusion structure.A SA attention mechanism is proposed to improve the performance of the improved network,and the ablation test is designed to verify the improvement effect.Results show that the file size of the improved model proposed in this paper is only 54.1 % of the size of the original network model,and the floating point operation of the network is only 26.87 % of the original network model,which reduces the requirement of the target detection network for the computing power of the load device while ensuring the target detection effect.(4)The Android smart phone platform is used to simulate the UAV software system platform,and the target detection network mobile terminal deployment test is carried out,and the effect is verified on the site of Zhushou Reservoir expansion project.Results show that the lightweight target detection network constructed in this paper runs smoothly on the mobile terminal platform.Using the target detection method proposed in this paper,it can effectively detect and identify the unsafe behavior recognition and target detection in the engineering site and automatically mark the construction site personnel,vehicle information and the behavior of construction personnel not wearing safety protection equipment.(5)The terrain analysis is carried out by using the UAV remote sensing data of Zhushou Reservoir dam site area to verify the rationality of road technical parameters.The numerical simulation model is constructed by using the digital elevation model.Combined with the physical and mechanical parameters of the material obtained from the field geological survey test,the safety state of the road rock and soil mass is analyzed by means of numerical simulation.Results show that the pressure load applied to the road by the construction transport vehicle mainly affects the location of the road,and the influence on the slope of the road is not obvious.The on-site road meets the needs of safe transportation in the construction area.