Curcuma, a key ingredient in curry and a popular health supplement, has been subject to adulteration and fraudulent origin labeling. In this study, E-eye, Flash GC e-nose, and FT-NIR, combined with machine learning and multivariate algorithms, were employed for origin identification and quantitative prediction of curcuma constituents. The results indicated that E-eye performed poorly in origin classification, while Flash GC e-nose identified flavor markers distinguishing curcuma from different origins but lacked precise quantification. After processing the FT-NIR spectra with SNV, the accuracy of three machine learning models, including SVM, increased from 83.3 % to 100 %. Additionally, PLSR models for three constituents, including curcumin, achieved mean R2 values exceeding 0.99 in both training and prediction sets, demonstrating excellent linearity and predictive accuracy. Overall, the study demonstrated that FT-NIR combined with multivariate algorithms provides an effective and feasible method for rapid origin identification and quality assessment of curcuma.