Machine Learning-Driven Landscape Perception Modeling: Combining the Visual and Acoustic Scene Analysis Methods of Segformer

In recent years, growing attention has been paid to how people perceive urban environments, particularly in the context of livability and intelligent city design. However, traditional perception research has often been limited by reliance on single-modal data or manual assessments. This study introduces a multimodal approach to modeling landscape perception by integrating both visual and acoustic scene analysis. Specifically, Segformer is used for extracting semantic information from street view images, while environmental audio is processed using a GRU-based emotion recognition model. These features are fused using a cross-attention mechanism to enhance perceptual prediction. We construct a dataset of paired street images and environmental sounds to evaluate the model. Experimental results show that the proposed method achieves more accurate and stable predictions than existing approaches, offering practical value for urban planning and public space evaluation.