Visual quality evaluation model of an urban river landscape based on random forest

• Main Authors: Dong, Y. (Author), Han, S. (Author), Li, L. (Author), Li, X. (Author), Lin, Q. (Author), Wang, X. (Author), Wu, D. (Author)
• Format: Article
• Language: English
• Published: Elsevier B.V. 2021
• Subjects: Decision trees; Deep learning; High quality; landscape; Landscape architecture; On-water landscape; perception; policy making; Quality control; Quality evaluation models; Random forest; Random forests; river basin; River landscape; river water; Rivers; Semantics; tourism; tourism development; Urban river; Visibility; Vision; visual analysis; Visual perception; Visual qualities; Visual quality evaluation
• Online Access: View Fulltext in Publisher

 A high-quality on-water landscape can improve the quality of cities and promote tourism development. However, current research on urban rivers has primarily focused on the riverside perspective, whereas few studies investigated the visual quality from an on-water perspective or conducted quantitative evaluations. This paper established a quantitative landscape index system by using a deep learning based semantic segmentation model to analyze human visual perception. A random forest model was used to analyze the nonlinear correlation between quantitative indicators and public scores, and an analysis and prediction model suitable for assessing the visual quality of an urban river on-water landscape was developed. This model provided high prediction accuracy and could rank the importance of the impact factors. The urban construction level, destructive index, hard revetment visibility, and green visibility index substantially affected the visual quality of the on-water landscape. The green visibility index was positively correlated, and the other three factors were negatively correlated with the visual quality. This model represents an intelligent approach for evaluating the visual perception and visual quality of the on-water landscape, enabling researchers and policymakers to analyze waterscapes from a new perspective and with high efficiency. © 2021 The Authors