OBJECTIVE To construct a scoring card capable of distinguishing between wild, semi-wild, and cultivated Astragali Radix. METHODS Six hundred batches of virtual data generated by TVAE deep learning were used as the training and validation sets. The training set data were binned, and the bins were adjusted and optimized to calculate the weight of evidence (WOE) for each bin. The data were encoded using WOE, and a logistic regression model was established. The model was trained on the training set and tuned using the validation set. A score of 50 was set as the threshold where the sample's positive and negative classification probabilities are equal. The baseline score and the scores corresponding to each bin were calculated using the formula and the logistic regression model equation. The sample score was determined by adding the base score and the scores corresponding to each bin for the sample, with a threshold of 50 used to judge the probability of the sample's positive or negative category. Card A and Card B were constructed to discriminate whether the Astragali Radix sample is wild or cultivated, respectively. RESULTS Sixty-four batches of real sample data were used as the test set to evaluate Card A and Card B, with classification accuracy rates of 0.86 and 0.80, respectively. CONCLUSION The scoring card can accurately discriminate the source of Astragali Radix samples, and the model is stable, reliable, easy to operate, and easy to promote.