Concrete moisture content testing is crucial in ensuring concrete quality, controlling the construction process, preventing cracks and deformations, and complying with industry standards. This paper proposes a near-infrared spectral concrete moisture inversion method that combines a hybrid model of Variational Mode Decomposition (VMD) and Long Short Term Memory (LSTM) to address the problem of rapid and real-time large-scale detection of concrete moisture content. Firstly, establish a standard concrete sample library with different moisture contents and complete near-infrared spectroscopy data collection under the same experimental conditions. Secondly, the raw near-infrared spectra are subjected to preprocessing such as smoothing, noise reduction, and standardization to analyze the near-infrared reflectance spectra produced by concrete with different moisture contents. Then, based on the collected near-infrared spectroscopy database, a VMD-LSTM hybrid model database is constructed, and the water content inversion performance of the hybrid model is evaluated by the root mean square error and determination coefficient of the evaluation indicators. Finally, the effectiveness of the method was verified through independent sample testing experiments. In summary, this study has achieved the advantages of rapid, non-destructive, non-invasive, and large-scale on-site detection of concrete moisture content inversion, which is of great significance for construction projects' safety and durability protection.