| With the development of the times,my country’s road network is rapidly expanding.By the end of 2019,the road paved mileage in my country has exceeded 5 million kilometers.However,with the expansion of roads,traffic accidents caused by road conditions(water accumulation,icing,dryness)are also on the rise,but my country still has shortcomings in the detection technology of road conditions,and most of the detection devices currently used on highways are from foreign countries,and the prices are relatively expensive.Therefore,this paper proposes a road condition detection scheme based on VNIR(visible-near infrared)spectroscopy technology and machine learning.The main research contents of the scheme are as follows:(1)The detection principle of the scheme was analyzed,the feasibility of the scheme was determined in principle,and the analysis process and detection methods of the nearinfrared spectroscopy analysis technology were introduced and analyzed.After comparative analysis,the near-infrared diffuse reflection detection method was finally used to obtain the road surface Spectral data.(2)The composition of the visible-near-infrared micro-spectrometer is introduced,and test experiments are designed for the integration time and repeatability of the instrument.The best integration time during day and night is determined,and the repeatability performance of the micro-spectrometer is verified.(3)According to the near-infrared diffuse reflectance detection method,a road reflectance spectrum collection system was designed,and the collection scheme was elaborated.The problem of instrument output saturation and the spectrum-like phenomenon of different roads in the daytime road spectrum collection were analyzed,and pointed the best working range of the instrument is 1/3～2/3 of the maximum output range.The method of installing a polarizer is proposed to solve the problem of output saturation of the instrument,and the obvious characteristic points of the reference background are added to solve the spectrum-like phenomenon..(4)A software platform for road condition detection is designed.Including the design of a user interactive host computer based on LabVIEW,which integrates the functions of spectral data acquisition,preprocessing,storage and instrument integration time control.The USB driver is written based on embedded C language to realize the USB transmission of spectral data.In order to realize the long-distance transmission of data,the network transmission program is designed using the integrated network port of the embedded development board S3C2440,so that the data can be transmitted to the remote terminal.And use the standard deviation calculation formula and the comparative analysis before and after data transmission to complete the performance verification of USB transmission and network transmission..(5)Establish a visible-near-infrared spectroscopy road condition recognition model.Before modeling,the spectral data was preprocessed by the selection of characteristic bands and smoothing and denoising.Because it is necessary to complete the detection of road conditions throughout the day,the recognition models for day and night were established respectively.Daytime model establishment part: In order to complete the distinction of various pavement conditions and the prediction of the thickness of the water pavement,the qualitative and quantitative analysis models of the pavement are successively established.This article uses SVM and BP neural network to establish the qualitative model,and compares the analysis.The experiment shows The qualitative model of BP neural network has better recognition effect,and the recognition rate of the verification set is 96.85%.After identifying the water pavement,a BP neural network with two hidden layers is established to qualitatively analyze the thickness of the water.The result shows the accuracy of the verification set.It is 86.56%.Night model establishment part: because there is no daytime spectrum phenomenon at night,there is no need to add additional feature points,so this paper establishes a BP neural network to complete the qualitative analysis of the road surface and the quantitative analysis of the water pavement at the same time.The experimental results show that the qualitative analysis is verified.The accuracy rate of the set is 100%,and the accuracy rate of the verification set in the quantitative analysis is 96.67%. |
