Study On Pavement Condition Recognition Based On Visible-near Infrared Spectroscopy

China’s highway traffic mileage ranks the first in the world and grows at a speed of about 100,000 kilometers per year.However,road condition detection devices in China are rarely used.In order to reduce the occurring rate of traffic accidents in severe weather conditions,China’s highway condition identification technology(whether there is ice,water,snow or dry)needs to be improved.Therefore,It was proposed a study on road condition recognition based on visible-near infrared spectroscopy that mainly used sunlight as the light source in this paper.The main contents include:(1)The calibration and performance test of miniature spectrometer were completed.Because the main control panel of the micro-spectrometer was self-made for the C11708 MA detector,it was calibrated with the standard spectrometer OSM-400.And the dark current,repeatability and integration time were determined by experiments.(2)An experimental platform for road condition recognition was designed.It was analyzed that the phenomenon of "Different substances with similar spectrum" in the state of ice and water during the day.And the method of light intensity characteristic value was used to solve the problem according to the characteristics of the phenomenon.The near infrared diffuse reflectance pavement detection device was designed,and the problem of light intensity signal saturation of micro-spectrometer was solved by means of polarizer device.The importance of adding the visible wave band to the near infrared wave band was explained.During the daytime experiment,sunlight was used as the light source,compared with the traditional light source(laser),it avoided the influence of sunlight on the traditional light source,and simplified the equipment and installation requirements.(3)The software platform was built.Experimental data was collected based on Lab VIEW host computer.The USB driver and network transmission software based on Linux were designed to realize long-distance data transmission,and the performance of USB drive was verified by standard deviation.(4)The road condition recognition was studied.The collection and study of experimental samples were divided into two parts of day and night.A total of 3230 and 240 road samples were collected under the four conditions of stagnant water,icing,snow and dryness,respectively.During the daytime,the sunlight was used as the light source,and the number of peaks and roughs,reflectivity range of the spectral curve of reflectivity could be used to identify the snow cover and dry state effectively.At the same time,logistic regression model,BP neural network model and BP neural network model based on Dropout were established to identify the conditions of icing and water accumulation.Through comparison and verification,the BP neural network model based on Dropout had a better effect and the recognition rate reached 98.21%.At night,the halogen tungsten lamp was used as the auxiliary light source.Since the phenomenon of "Different substances with similar spectruml" will not occur in freezing and water accumulation at night,the method of "combination threshold value" could be used to identify the road condition,and could effectively distinguish each condition.In summary,the road condition recognition technology based on visible-near-infrared spectroscopy proposed in this paper can effectively realize the recognition of road icing,water,snow and dry conditions,which provides an effective means for road condition recognition.