GazeNet: Neural Network-Based Visual Attention Simulation for Museum Exhibition Optimization

Effective visual communication in cultural heritage exhibitions requires strategic organization of information elements to guide visitor attention and enhance learning outcomes. Current exhibition design methodologies depend on empirical knowledge and resource-intensive user testing, creating barriers to evidence-based optimization. This research presents a comprehensive deep learning solution that automates visual attention analysis for exhibition layout assessment and improvement. The proposed system combines residual neural networks with transformer-based attention modules to model human visual scanning behavior, providing quantitative metrics for design evaluation without requiring participant studies. The architecture processes exhibition imagery through hierarchical feature extraction while employing cross-attention mechanisms to predict visitor gaze sequences and fixation distributions across display elements. Experimental results on a diverse dataset of museum displays demonstrate that our method achieves superior gaze prediction accuracy compared to state-of-the-art approaches. Comprehensive evaluations across multiple exhibition types reveal the framework’s effectiveness in identifying design deficiencies, quantifying visual attention distribution, and generating actionable optimization recommendations. This computational approach empowers cultural heritage practitioners to systematically evaluate and refine exhibition designs, improving information accessibility and visitor engagement through optimized visual hierarchies. The technology offers practical applications for museum professionals, exhibition designers, and cultural education specialists seeking to enhance knowledge transmission effectiveness while reducing design iteration costs.