In this study, 110 dried Panax notoginseng were selected as the test samples. Computer vision technology was used to obtain the images of Panax notoginseng taproot, which were deal with gray, binary and morphological to extract the length, width and projection area was ed after preprocess.The prediction models were built to calculate the projection area and the weight for cone Panax notoginseng and nodule Panax notoginseng, respectively. The results showed that the weight of mainroot was linely correlated with the projection area. The determination coefficients of cone Panax notoginseng and nodule Panax notoginseng were 0.984 9 and 0.986 6, respectively. The quality prediction model was verified by 10-fold cross-validation method. The average error was 0.3348 g and 0.494 9 g for cone Panax notoginseng and nodule Panax notoginseng, respectively.