To excavate users'emotional preferences more accurately,an image-driven intelligent design method of product modeling was proposed using artificial intelligence technology to assist designing product solutions to meet the requirements.The product online review data was filtered.Term Frequency-Inverse Document Frequence(TF-IDF)technique was applied to extract the representative perceptual image vocabulary that described product modeling.The target image was obtained after cluster analysis.The sample image evaluation value was obtained combined with the semantic difference scale.GoogLeNet convolutional neural network was used to construct the image regression model.The scores of other samples were predicted to obtain the image evaluation data.The constituent elements of the sample were deconstructed.The prompt words were trained according to the importance ranking setting.Finally,the Low-Rank Adaption(LoRA)image modeling generation model was constructed by fine-tuning the Stable Diffusion XL(SDXL).Taking the hairdryer as an example,the innovation design scheme of the target perceptual image was generated.The feasibility and rationality of the method were verified,that could effectively guide the product innovation design.
NETL
NSTL
万方数据
