Analysis And Evaluation Of Urban Street Space Quality Change Based On Internet Street View Images

It is not only a place for public activities in the city,but also an important carrier for people to express themselves and express their emotions.However,in recent years,along with the accelerated pace of urbanisation in China,a car-oriented urban road planning has been formed-with vehicles taking up too much space,poor walking environment,people feeling inconvenienced and unsafe when walking on the streets,insufficient greenery and poor environmental quality are serious and particularly prominent problems.With the advantages of wide coverage,low acquisition cost and fitting human perspective,Internet streetscape images provide new data sources for research in various fields of urban planning,construction and management.Therefore,this paper takes urban streets as the research object,takes Internet street view images as the data source,uses deep learning technology to carry out automatic identification of street space quality elements,constructs a scientific,comprehensive and effective street space quality evaluation model,evaluates and analyzes the spatial quality of urban streets,explores its spatial quality change trends,provides auxiliary data support for the formulation of urban street design guidelines and quality improvement,and helps improve the sense of existence and satisfaction of urban residents.It helps to improve the sense of existence and satisfaction of urban residents.The specific research contents and results are as follows:(1)Firstly,the spatial quality elements of streets in streetscape images were extracted.Based on the historical data of Baidu Street View images,OSM urban road network data and Cityscapes dataset(including 5000 refined annotated images),a deep learning PSPnet network is used to construct a semantic feature recognition model for street spatial elements.Among them,MIou value and pixel accuracy Accuracy are used as the evaluation indexes of the recognition model,and the pixel occupancy statistics formula is used to quantify the street spatial quality components.(2)Then the evaluation model of street space quality was constructed.Based on the current situation of street space quality research at home and abroad,the evaluation indexes and evaluation methods of street space quality in different regions are summarized,the subjective(5 items)and objective evaluation indexes(4 items)affecting street space quality are selected,and their scoring and grading methods are determined,and the evaluation model of urban street space quality is constructed by combining the mainstream hierarchical analysis method.(3)Finally,an example validation of the research method is carried out.A total of 22 streets with 449 sampling points and 4332 Baidu street images are selected for feature extraction and analysis in the Nanluoguxiang neighborhood of Beijing.In the recognition results of the semantic feature recognition model of spatial elements of streets,the MIou value of the training set is 0.80 and the Accuracy value is 0.96;the MIou value of the validation set is 0.70 and the Accuracy value is 0.96,which meet the requirements of element recognition accuracy.The top three categories of spatial elements in the study area were roads,buildings and vegetation,accounting for 75.96% of all feature elements.The results of street quality evaluation show that the number of sample points with high spatial quality in the study area only accounts for 12% of the total;among the roads,Di’anmen East Street has the highest spatial quality and Qin Lao Hutong has the lowest spatial quality.Combined with the streetscape images in different periods,we found that among the arterial roads,Di’anmen East Street had the highest spatial quality change(6.47%)and Gulou East Street had the lowest spatial quality change(2.72%);among the hutongs,Shajing Hutong had the highest spatial quality change(6.07%)and Doujiao Hutong had the lowest change(3.70%).