Application and Optimization Strategies of the Improved ResNet50 Model in Image Aesthetic Scoring

Abstract

This research introduces an image aesthetics evaluation model designed to enhance the accuracy and robustness of image aesthetics scoring by modifying the pre-trained ResNet50 model. The experimental findings indicate that this model surpasses other prevalent convolutional neural network architectures and optimizer combinations. Additionally, this paper examines the optimal configuration of ResNet50 combined with the Adam optimization algorithm with a learning rate set at 0.001. A thorough analysis reveals that the residual block design of ResNet50 is particularly effective in managing deep features, while the Adam optimizer offers substantial benefits in maintaining stable convergence and efficient learning. The outcomes suggest that this model has extensive application potential in various scenarios, including social media platforms, image retrieval and search engines, photography training and automated design, advertising creation and content generation, smart albums and image management, and artistic creation and review. In these applications, the model can substantially improve the precision and performance of image recommendation, retrieval, classification, and evaluation, resulting in a superior user experience.