The Application of Artificial Intelligence Tools Based on Convolutional Neural Networks in Creative Design

Creative design has made significant progress with technological changes, and artificial intelligence tools, especially models based on Convolutional Neural Networks (CNN), are gradually becoming designers' new partners. This study achieved the extraction and representation of design features by constructing a CNN model. This model extracts local features of the image and reduces the dimensionality of the feature space through ReLU activation function and max pooling operation. The generalization ability and training efficiency of the model are improved through L2 regularization and batch normalization techniques. The research results found that compared with traditional design methods, CNN models perform better in terms of CPU usage and design originality. Experimental data shows that the average CPU usage of CNN in creative design tasks is 23.808%, which is much lower than that of Recurrent Neural Networks (RNNs) and Autoencoders. In addition, the originality of creative design achieved by CNN is as high as 98.4%, significantly higher than that of RNN and Autoencoder. However, these tools still face challenges in controlling accuracy and copyright issues. Future research needs to further optimize CNN models to improve the stability and diversity of generated content.