Research And Application Of Fast Image Analysis On Pore Structure Of Concrete Based On Deep Learning

Concrete materials are multi-phase,heterogeneous and multi-layer composite systems with complex structures.The macroscopic physical and mechanical behaviors exhibit irregularities,uncertainties,nonlinearities,etc.The reflection.Cement-based materials should be studied from three levels: microscopic,mesoscopic and macroscopic,and should be further studied in microscopic research.As an important part of the microscopic and mesoscopic layers of concrete materials,the pore structure has an important influence on the macroscopic properties of concrete.In recent years,more scholars have studied the pore structure of concrete.Image analysis method is a method to study the structural parameters of pores based on image.However,the existing image analysis method has some problems,like the accuracy,the complicated operation process,and the need of professional operators.In order to solve the problem of image analysis method,this paper applies the deep learning technology to concrete pore structure analysis for the first time,which improves the accuracy and efficiency of image analysis method,and simplifies the analysis process.Through the deep learning algorithm,a deep learning model for accurately and efficiently identifying pore structures of concrete with apertures larger than 10?m was established.In this paper,a data containing 8000 pore structure images concrete and mortar was seted.The minimum sample of deep learning image analysis is determined by statistical theory.The section of 40 mm×40 mm mortar needs 80 pieces,and the section of 100 mm×100 mm concrete needs 150 pieces.Combining with professional image software,the quantitative analysis of pore structure parameters of concrete is carried out,and a fast method of pore structure of concrete based on deep learning is established.The feasibility of this method is verified by analyzing the relationship between fractal dimension of pore and strength of test block.The method for fast analysis on pore structure of concrete based on deep learning has an error rate of 0.8% and 1.5% for the pore structure of mortar and concrete on the test set,and the accuracy rate is higher than 98%.The error rate is 7.8%,on test mortar and the accuracy rate is higher than 90%.Compared with traditional image analysis and different deep learning algorithms,the accuracy of method of this paper is higher.In terms of efficiency,this method avoids the pretreatment of cross-section,simplifies the experimental process,reduces the difficulty of the experiment,and avoids the selection of image threshold.Compared with traditional image analysis methods and different deep learning algorithms,the time of recognizing a single image is shorter.