Objective: To realize rapid nondestructive testing of rice freshness. Methods: A portable device system based on nano color-sensitive sensors combined with near-infrared spectroscopy was developed. The spectral data of the corresponding color-sensitive sensors were collected for the identification of multi-gradient rice adulteration and the prediction of freshness across batches of rice. Results: Using Si-CARS-PLS to extract spectral characteristic variables, the recognition rate of discriminative model was the highest after modeling by LDA algorithm, and the recognition rate of training set and prediction set were 97.22% and 95.83%, respectively. At the same time, PLSR model predicted that cross-batch data had stronger stability. The correlation coefficients （R_c, R_p） of the training set and the prediction set of different batches of rice sample data were all stable at about 0.95, the root mean square errors （RMSEC, RMSEP） were all lower than 0.2, and the relative analysis errors （RPD） were all greater than 3. Conclusion: The system has the characteristics of high accuracy, convenience and good robustness of prediction model, and has a good application prospect in the field detection of rice freshness.