The compactness of roadbeds is an important factor affecting highway construction’s quality and service. Therefore, it is of great practical demand and significance to grasp the compactness of highway subgrade quickly, non-destructive and accurately. However, the traditional detection of the compactness of highway roadbeds is mainly based on the accurate detection of a small number of discrete points, which can not meet the need for comprehensive and accurate detection of roadbed construction quality. Hyperspectral technology is a high and new technology that can realize real-time, fast, non-destructive and accurate monitoring of surface information, providing a new solution for detecting compactness of highway roadbeds. In order to explore the feasibility of hyperspectral technology in detecting the soil compaction degree of highway roadbed, soil compaction degree and corresponding spectral data were obtained through the soil compaction experiment and soil spectral measurement experiment, and the soil compactness coefficient was constructed with the help of soil spectral response mechanism analysis. Then the soil compactness coefficient is constructed by soil spectrum before and after compaction, and the soil compactness coefficient is processed and analyzed by discrete wavelet algorithm. The correlation between low frequency and high-frequency information and soil maximum dry density is analyzed quantitatively by correlation algorithm. The characteristic bands are extracted and screened, and then the soil maximum dry density estimation model is constructed based on the partial least square algorithm. The results show that: (1) after compaction, the soil spectrum decreases with the increase of soil water content, and the decrease range increases with the increase of soil water content, and the relationship between the variation range of soil spectral reflectance and the difference of soil water content is non-linear. Compared with the soil spectrum before compaction, except for 20% soil water content, the soil spectral reflectance increases or decreases in different degrees in the whole band range after compaction, and this change is easy to have a certain impact on the detection of soil composition. (2) the soil compactness coefficient generated by the soil spectrum before and after compaction can improve the sensitivity of the spectrum to the maximum soil dry density after compaction, and the correlation coefficient R is up to 0.811, which is highly correlated. (3) under the discrete wavelet algorithm, high-frequency information can improve the ability of soil compactness to estimate the maximum dry density of soil, among which the model based on D1 has the highest accuracy and is the best model, and its R2=0.957 and RMSE=0.023. The resolution of the soil compactness coefficient greatly influences the accuracy of the estimation model. The research results of this paper can provide basic theory and method support for the application of hyperspectral technology to the monitoring of highway roadbed compaction, other engineering foundation compaction and topsoil compactness.