At present, the vibration nondestructive testing method has high requirements on hardware system and test conditions, and few practical application systems have been designed..In order to solve this problem, a portable vibration test detection system was built. The time signal was denoised by the method of wavelet threshold denoising. The sub-band spectrum centroid of a certain watermelon variety was extracted. The product of square of transverse diameter and the longitudinal diameter was proposed as the volume parameters. It is found that the frequency point of sub-band spectrum centroid is negatively correlated with the quality of watermelon and the volume parameters, and is less dependent on the intensity of the excitation. Using stepwise multivariate linear regression analysis and surface modeling established the sugar detection model, and the quality of watermelon, the frequency point of sub-band spectrum centroid and the volume parameters were variables. The model established by surface modeling method is better. The determination coefficient of the model reached 0.881 1, the standard deviation of the prediction set sample RMSEP was slightly larger than the standard deviation of the calibration set RMSEC, and RMSEP and RMSEC were both less than 10% of the maximum sugar content. The model can be well used to predict the sugar content of the watermelons of the maturity of 80% to 100% with the frequency point of sub-band spectrum centroid of 216 to 268 Hz.