犯罪恐惧感的空间格局、影响因素及其与犯罪活动的异同——基于街景图像深度学习的北京案例

Spatial pattern and influencing factors of fear of crime and their differences with those of criminal activities:Application of deep learning algorithm to street view images in Beijing

犯罪地理研究始终存在重视客观犯罪、轻视主观感知的缺憾。论文以北京为案例,采用深度学习中的图像回归算法,大规模识别不同街景环境下的犯罪恐惧感,以弥补社会调查在空间覆盖率、空间分辨率以及信度效度等方面的不足。研究表明:①犯罪恐惧感的空间格局呈现出由中心向郊区逐渐递增的圈层式、多组团、放射状结构,盗窃及暴力犯罪量的空间分布与之大相径庭。②由犯罪恐惧感与犯罪活动的空间分布匹配关系可见,在北京市中心,主客观安全性具有低匹配度且客观形势比主观感知更危险;至近郊,两者匹配度提高;至远郊,两者仍具低匹配度但客观形势比主观感知更安全。③主客观安全性的影响因素不尽相同。高密度、混合型环境有利于降低恐惧感却会诱发犯罪;增设尽端路、提升围合感、增加绿视率、整治物理失序则能发挥抑制恐惧感和犯罪量的双重功效;弱势社区兼具高恐惧感与高暴力犯罪量,人员流动性有助弱化恐惧感,居民异质性仅会加剧犯罪行为。研究结果有助于厘清经典犯罪地理理论对主客观安全性的差异化解释力,以全面评估环境干预政策的安全性后果。

Research on crime geography has traditionally focused on objective criminal activities and neglected subjective perceptions such as fear of crime,which is its major shortcoming.Using the deep learning algorithm of image regression,we analyzed the level of fear of crime under different streetscape environments on a large scale in the central urban area of Beijing.This approach compensates for the limitations of social surveys in terms of spatial coverage,spatial resolution,and reliability and validity of measures.Our indigenous deep learning models also make up for the lack of established models that rely on western city street view images and overseas labelers.Our study shows that,first,the spatial pattern of fear of crime has a circular,multi-cluster,and radial structure,and its level gradually increases from the city center to the suburbs.In contrast,the density/number of theft and violent crimes have the spatial distribution with the opposite trend.Second,according to the relationship between the spatial distribution of fear of crime and criminal activities in general,we find a low match between subjective and objective security in the city center,with the objective situation being more dangerous than the subjective perception;in the suburban areas,the degree of match between the two increases;to the outer suburbs,these two still have a low degree of match,but the objective situation is safer than the subjective perception.Third,the built and social environmental factors that influence subjective and objective security are not always identical.A high-density and highly mixed environment would reduce fear of crime,but may accelerate crime.To reduce both fear and crime,it is recommended to add cul-de-sacs,improve the sense of enclosure,increase the amount of greenery,and eliminate various types of physical disorder.As for the effect of social disorganization,concentrated disadvantaged communities tend to have high levels of fear and are more prone to violent crime;population mobility can help reduce fear of crime;residential heterogeneity would exacerbate criminal behavior.Our findings help to clarify the differential explanatory power of classical crime geography theory for subjective and objective security,which in turn facilitates a comprehensive assessment of the security consequences of environmental intervention policies.