图像深度学习技术支持下南京秦淮河滨水景观视觉质量评价研究

Research on the Visual Landscape Quality Assessment of Qinhuai River Waterfront Landscape in Nanjing with the Support of Image Deep Learning Technology

将南京秦淮河线性滨水景观选作研究对象,运用主客观感知要素相互结合的方式对城市滨水景观视觉质量评价展开定量研究。通过设计一种利用远程云端技术收集数据的方法,将自摄照片集作为图像机器学习载体,训练出一套适用于城市滨水景观的图像语义分割模型与情感拟合算法,构建起一套可以量化解析客观视觉特征与主观视觉感知之间相关性的关联模型。最后,通过K折交叉验证及相关模型进行自验与他验,检验视觉感知关联模型的准确性与适用性。结果显示:图像深度学习技术可以较好地模拟人类视觉感知;将视觉感知关联量化模型应用于实际场景,能够预测人类对于大尺度滨水景观的视觉感知结果;从而回应城市滨水景观视觉质量评价的应用需求与价值导向,为城市滨水景观资源管理提供合理建议。

The linear waterfront landscape of Qinhuai River in Nanjing is selected as the research object to conduct quantitative research on the visual quality assessment of urban waterfront landscape by combining subjective and objective perception elements.By designing a method of data collection using remote cloud technology and using self-shot photography sets as image machine learning carriers,a set of image semantic segmentation model and emotion fitting algorithm applicable to urban waterfront landscape are trained,and a set of correlation model can be constructed to quantitatively analyze the correlation between objective visual features and subjective visual perception.Finally,the accuracy and applicability of the visual perception association model are verified by self-test and other tests through the K-fold crossover algorithm and comparison with other quantitative model.The results show that the image deep learning technique can better simulate human visual perception.By applying the visual perceptual correlation quantification model to the actual scenario,it can predict the results of human visual perception of large-scale waterfront landscape;thus,it can respond to the application needs and value orientation of the visual landscape quality evaluation of urban waterfront environment and provide reasonable suggestions for the management of urban waterfront environment visual landscape resources.