In response to the problems of insufficient personalization and unclear featureextraction in traditional visual image design, this paper studied an image feature extraction and processing method based on CNN (Convolu...
详细信息
ISBN:
(纸本)9798400718144
In response to the problems of insufficient personalization and unclear featureextraction in traditional visual image design, this paper studied an image feature extraction and processing method based on CNN (Convolutional Neural Networks). Firstly, the images were preprocessed and enhanced to enhance the generalization ability and featureextraction performance of the CNN model;a lightweight CNN architecture and MobileNet (Mobile Network) were designed to reduce computational complexity and parameters, and the model was optimized through technical means;generative adversarial networks (GANs) were utilized to integrate creative elements and enhance the innovation of advertising design. Through transfer learning, the CNN model was adjusted, and the Adam optimizer and early stop strategy were adopted for model training and optimization. In the experiment of featureextraction accuracy, the accuracy of the model in this paper reached 98.8%, 97.8%, and 98%, and it was also better than the comparison model in the comparative experiment. In the generalization ability test, the model did not overfit and demonstrated good generalization ability;in the processing time test, the architecture used in this paper had a shorter processing time than other architectures;in the survey on the integration effect of creative elements, the attractiveness rating reached 96% of the positive reviews.
暂无评论