Traditional Chinese medicines (TCMs) are in great demands around the world, and their popularity increased even more when Tu Youyou was awarded the 2015 Nobel Prize in Medicine. However, many key issues regarding the current quality standards remain and moisture content (MC) of TCM is one of the important elements of the quality control inspection. MC detection of TCM is usually implemented by using a voltage level or lock-in amplifier based on a Near-infrared Light-Emitting Diode (NIR-LED) absorption spectroscopy sensor. Square signal is frequently used as excitation source due to its abundant absorption features both in the time-domain and frequency-domain. However, the planeness, lift-off effect, and measurement noise negatively affect the time-domain features during MC measurement. A MC sensor has been developed to precisely measure the MC of Dutong pills based on NIR-LED spectrum analysis in this work. Firstly, a broadband NIR-LED with a central wavelength at 1 450 nm was adopted for the water absorption source. The spectrum features of the direct absorption signals were extracted by Fast Fourier Transform (FFT) algorithm. Secondly, the parameters of the excitation square signal were analyzed and optimized. The optimization results showed that the direct absorption spectrum signal was optimal when the excitation frequency, amplitude and duty ratio were set as 100 Hz, 1500 mV and 50% respectively. Finally, eight calibration groups with different MCs and four validation groups of Dutong pills were prepared. The linear relationship between the MC of the Dutong pills and the odd harmonic signal amplitude was established on the basis of the optimized square parameters of the excitation source. The calibration and validation results demonstrated the fundamental frequency components (100 Hz) and the 3rd harmonic components (300 Hz) have a high linear correlation with the MCs of the Dutong pills and the linear correlation coefficient R values were 0.992 and 0.993, respectively. The minimum measurement error and maximum error after averaging were 1.0% and less than 6.5%, respectively. The MC sensor took the square wave as excitation source. The noise of the water absorption spectrum signal was removed by wavelet filtering algorithm. The FFT algorithm was adopted to extract the feature of spectral signal. Therefore, the sensor has the advantages of high precision, wide applicability, real-time online measurement, and has the potential to be used for quality control and process analytical technology in TCM.