Spectral technology is a promising prospect for highway state detection(whether frozen, water accumulated, or snow accumulated). However, there is little research on using sunlight as a light source to identify highway states. Sunlight and halogen tungsten lamps were used as experimental light sources in the day and night. Spectral curves of the visible-near-infrared bands of ice, water, snow, and highway backgrounds were obtained using a micro-spectrometer. During the day, the state of icing and stagnant water resulted in a phenomenon known as "Different substances with similar spectra" under different illumination conditions. Then, based on the characteristics of sunlight illumination, the solution of "environmental variables" as eigen values was proposed. The curve of the spectrum and the normalized "environmental variables" were combined into a new data waveform, and a neural network model based on Dropout and an Adam optimizer was established for training and recognition. The final recognition rate was 99.375%. At night, due to the evident differences in the spectra of various samples, the spectral curves of each sample were identified using the "combination-threshold" method. Experiments proved that the method of combining two light sources can effectively identify the road surface state.