The work aims to develop a product CMF decision model by integrating BP neural network and linear regression in order to accomplish accurate selection and quantification of product CMF under certain perceptual images. Through text mining, the user's perceptual image was identified, a CMF element space was created according to the HSV color model and the material and process of chosen rehabilitation aids, a large number of CMF solutions based on the design element space were formed, and the solutions were assessed in accordance with the chosen perceptual image to obtain a qualitative mapping relationship between the perceptual image and a single design element of CMF. The CMF solutions were coded and integrated with the perceptual image evaluation value and the CMF decision model was established by quantitative method through BP neutral network to identify the best color space, material and method. The chosen color intervals were separated into design solutions and evaluated. The color regression equation was then created by linear regression, and the CMF decision model was quantitatively built by BP neural network. With knee brace as an example, the case study was carried out, the first-order CMF decision model developed by BP neural network had an MSE of 0.038 13 between the predicted and expected values, and the prediction results were essentially consistent with the qualitative mapping relationship, demonstrating the high accuracy and reliability of this order of the model. The values of H, S, and V were highly associated with the perceptual image assessment value, as shown by the second-order decision model's p-value of less than 0.01, which demonstrated the viability of the CMF decision model. The developed CMF decision model has some versatility in the area of product design and is capable of realizing correct CMF selection and quantification for rehabilitation items as well as directing preferential CMF decision and innovation at both the qualitative and quantitative levels.