Research on the exhibit layout model driven by visual saliency in exhibition hall space with particle swarm optimization
Xianhui Zeng, Simeng Fan · Entertainment Computing · 2026
To optimise the layout of exhibition hall space and enhance the visual appeal of exhibits, this paper proposes an innovative exhibit layout model driven by visual saliency. This model deeply integrates the particle swarm optimization algorithm with a deep learning-based visual saliency calculation to achieve automated, intelligent exhibit layout through a data-driven approach. The visual saliency prediction model is used to analyze the panoramic view of the empty exhibition hall, generating a visual attention distribution map of the quantified space. An optimisation model with exhibit coordinates as particles is constructed, and its fitness function maximises the global visual benefit as its core objective while strictly satisfying spatial constraints, such as non-overlapping. The core innovation lies in improving the standard particle swarm algorithm by introducing a salience-guided initialisation strategy and an adaptive inertia-weight adjustment mechanism, enabling the optimisation process to converge to a visually optimal solution efficiently. The experimental results in the simulated exhibition hall environment show that, compared with the traditional empirical layout method, the overall visual appeal of the layout scheme generated by this model is increased by about 25% on average; compared with the standard particle swarm algorithm, the convergence speed of this model is increased by about 40%, and it can effectively avoid local optima. This research provides a scientific quantitative tool and a new solution for exhibition hall layout design, with important theoretical value and application prospects.