物理失序对盗窃犯罪的影响及其在3D建成环境影响中的调节作用——基于街景图像深度学习的上海案例

The impact of physical disorder on theft crime and its moderating role on the criminogenic effect of 3D built environment features:Application of deep learning approach on street view images in Shanghai

破窗理论是广受关注但缺乏本土检验的犯罪地理理论。本文以上海中心城区为例,利用目标检测、图像回归等深度学习算法挖掘街景图片中的具体失序实物与整体失序感知,以厘清物理失序对盗窃犯罪的影响及其在密度、混合度、设计要素影响盗窃分布过程中发挥的调节作用。研究表明:①盗窃犯罪呈中心高、外围低、多热点的空间格局,失序实物与感知的空间分布异中有同。堆放垃圾和脏乱差得分由中心向外围呈先降后升态势,侵占道路在市中心高发,涂鸦小广告相对分散,而部分城中村、棚户区面临多重失序叠加的困境。②除侵占道路外,其余物理失序对盗窃活动均有显著正向影响,整体失序感知的影响最强,强度仅次于可步行性,并与活动人口数相当。③物理失序在3D建成环境要素影响盗窃行为的过程中起着增强、削弱、干涉调节作用。物理失序加深会增强POI密度、商铺密度、路网密度、街景多样性、围墙领域感的原有影响,削弱POI多样性的正向影响,将空间开敞度、灌木绿视率等存在双刃剑效应的要素影响方向由负转正。总之,加强环境维护管理对抑制犯罪发生具有可行性、有效性、低成本等优势。

The broken window theory is one of the typical environmental criminology theories that have attracted considerable academic attention internationally,but lacks validation in the Chinese context.Taking the central urban area of Shanghai as an example,we use deep learning approaches,including object detection and image regression,to identify disordered physical objects and overall perception of disorder from street view images.We then examine the impact of physical disorder on theft crime and uncover its moderating effect in the process by which 3D built environmental features(i.e.,density,diversity,and design)influence theft crime.The results show that,first,the spatial pattern of theft crime exhibits a central-high,peripheral-low distribution with multiple hotspots.The spatial distribution of multiple disordered physical objects and the overall perception of disorder show both similarities and differences.Garbage piles and perception of untidiness show a decreasing and then increasing trend from the center to the periphery,street encroachment is high in the city center,graffiti and small advertisements are relatively dispersed,while some of the urban villages and shantytowns face the dilemma of multiple disturbances stacked on top of each other.Second,apart from street encroachment,all other physical disorder phenomena have a significant and positive direct impact on theft crime.The most prominent factor among them is the overall perception of untidiness,whose magnitude of influence is second only to walkability among all independent variables and covariates,and comparable to the impact of the size of ambient population.Third,physical disorder plays mediating roles in enhancing,diminishing,and interfering with the impact of 3D built environment features on theft crime.Increasing physical disorder would enhance the criminogenic effects of POI density,store density,street network density,streetscape diversity,and the sense of enclosure provided by fences,weaken the positive impact of Points of Interest(POI)diversity on theft,and reverse the direction of the relationship between environmental factors,such as sky openness and green view ratio of shrub,and theft from negative to positive.The study provides evidence that enhancing the policy related to environmental maintenance and routine management offers a practical and cost-effective way to inhibit the occurrence of theft crime.