Image style recognition and intelligent design of oiled paper bamboo umbrella based on deep learning

Abstract

The intelligent design of cultural and creative products has become the research hot spot in the field of computer aided design. Aiming at the disadvantages of low recognition accuracy and manual feature extraction of the early product image recognition model, this research uses the image recognition model for cultural and creative products constructed with deep convolutional neural network. In this research, the classical oiled paper umbrella is taken as the example, and an image recognition experiment on the umbrella has been carried out with a kind of Residual Network (ResNet-50) that is based on convolutional neural network. The experimental result shows that the ResNet-50 convolutional neural network is effective and accurate for product image recognition, with a accuracy rate reaching 94.3%. On the basis of the image recognition of oiled paper umbrella, an experiment has also been carried out on the conditional generative adversarial network (Deep Convolutional GAN, DCCAN), with both generator and discriminator adopt deep neural network. The input of generator is Gaussian noise, which generates a series of classical oiled paper umbrella through Deep Convolutional GAN, and the experimental result is the product creative design scheme. The experimental result shows that the sample images generated by conditional generative adversarial network are feasible and effective, and that the model can generate some shallow classical oiled paper umbrellas, which assists the inspiration design of oiled paper umbrella. This research provides a new idea for modeling of product image recognition, and overcomes the disadvantages of the traditional product design, such as too much reliance on designers, complicated process and low design efficiency.