Research On Intelligent Identification Of Hidden Dangers In Deep Foundation Pit Construction Process Based On Deep Learning

Deep foundation pit is one of the graveness hazard in the field of building construction.Its accidents are often accompanied by large losses,wide influence and high risk.It is found that many hidden dangers appear occur in the process of construction,and how to further ensure the construction safety has become an urgent problem to be solved.In recent years,with the rapid development of computer technology,the use of deep learning methods to identify risk factors efficiently and timely has become an important research direction.At present,the inspection method of deep foundation pit engineering mainly relies on manual work,which faces the problem of high inspection cost.At the same time,the application of image recognition in the field of construction safety is mostly concentrated in a single,common or not limited to the object of engineering scene.In view of the above problems,this paper focuses on the specific risk factors of deep foundation pit site,studies and puts forward the intelligent identification method of deep foundation pit construction hidden dangers.The main results are as follows :(1)The risk factors leading to deep foundation pit accidents often have regularity.Therefore,based on the survey report,this paper uses text mining technology to screen out some hidden danger factors of deep foundation pit.The key hidden dangers in the four types of construction process determined,including improper or missing construction procedures,inadequate protection of the edge,water accumulation and slope overload;(2)Based on YOLOv5 network,the detection of three kinds of hidden dangers,such as inadequate protection of foundation pit,water accumulation and slope overload,is better realized.Firstly,the Bi-directional Feature Pyramid Network feature fusion mechanism is used to improve the recognition accuracy.Secondly,on this basis,the model burden is reduced by means of channel pruning.Finally,different scaling factors and pruning rates are set to determine the model with the best performance,and the comparison and verification are carried out.The results shows that the accuracy of the model is increased by 1.4 %,and the FPS is increased by about 52.91.(3)In view of the hidden danger of improper or missing construction procedures,we propose two models: Based on the Slowfast network model,the recognition effect is guaranteed by replacing the original detector with YOLOv5.In addition,the Deep SORT module is added to track the progress of the process.Experiments show that the algorithm model can accurately track the process and detect the job content,and the accuracy rate is also improved by about 2.9 % compared with the original;In order to ensure that the construction is carried out according to the design process,a sequence identification network based on Unet is proposed.The experiment proves that the model can realize the sequential detection function,the accuracy rate is about 98 %,and the overall performance of the model is good.(4)The construction safety knowledge base is established by combining Neo4 j graph database and Python language to guide the safety and standardized operation of deep foundation pit engineers.It can finally provide query functions such as standard construction specifications,process flow and parameters and emergency measures.At the same time,an intelligent control platform for deep foundation pit construction is proposed,which provides a new perspective for the development of deep foundation pit management.