Dynamic Risk Assessment Of Urban Street Crime Coupling Multiple Human-environment Factors

Since the reform and opening up,various social development problems in cities are the profound reasons for urban public security risks.The urban security and stable development are facing severe challenges in China.The high-intensity risk prevention and control policy adopted by the management department has achieved remarkable and effective results while consuming a lot of manpower and material resources.It is urgent to improve the scientificity,accuracy,effectiveness of the social public security management system.As a major type of urban crime that occurs in public places,street crime has a wide range of social damage and is difficult to control.It seriously undermines urban social order and gravely damages the security of people’s lives and properties.Dynamic quantitative assessment of street crime risk that calculate the potential probability distribution of crime events in geographical space can provide a strong scientific basis for the rational distribution and optimization of police resources.And it can effectively improve the ability of public security departments to accurately and efficiently combat and prevent crime,which is also an important research direction in the field of crime geography.Enabled by new technologies such as space-air-ground collaborative observation and ubiquitous sensing interconnection,multi-source urban spatial big data provides unprecedented opportunities for large-scale,fine-grained dynamic risk assessment of urban street crime.Therefore,from the perspective of human-environment interaction,this thesis uses street view images,human mobility,social economy and other urban perception big data to carry out the theory and method research of dynamic risk assessment of urban street crime with the help of spatial statistical learning and artificial intelligence technology,including:First,aiming at the environmental effect of urban built environment on street crime risk,a quantitative evaluation method of street crime risk considering the effects of pairwise variable interactions is proposed.Firstly,we used a deep neural network to extract multi-elements of largescale urban built environment from street view images.Then,constructed the multi-view environmental vectors combining the socioeconomic and demographic information.And a zero-inflated negative binomial regression model is established by fusing the multiplication-based interaction effect.It effectively quantifies the synergistic effect of multiple built environment variables on urban street crime risk.The experimental results show that compared with independent single variables,the effects of pairwise variable interactions can more accurately and deeply explain the quantitative impact of urban built environment elements on street crime risk.Second,aiming at the effect of human travel activities on the street crime risk,a quantitative evaluation method of street crime risk change considering the difference of spatiotemporal distribution of human travel is constructed.Firstly,the non-negative matrix factorization model with spatiotemporal constraints is used to automatically extract the change pattern of human travel.Then,with the help of zero-inflated negative binomial regression model,we effectively reveals the significant correlation between the spatiotemporal changes of human travel and the changes of street crime risk.With the implementation of COVID-19 travel restriction policy is taken as the research background,we combined with the semantics of human travel activities,a reasonable explanation is given for the spatial differences of street crime risk changes.Third,aiming at the multi-factor comprehensive effects of human travel activities and geographical environment on the street crime risk,a multi-scale dynamic risk assessment method of street crime is developed.Firstly,the zero-inflated phenomenon is alleviated by enhancing the original data,and a time-varying similarity matrix is constructed to measure the spatiotemporal correlation intensity of crime risk.Then,a deep neural network integrating graph convolution and sequence learning is constructed,and a cross-scale joint decoding strategy is used to invert the spatiotemporal evolution trend of street crime risk.The experimental results show that this method significantly improves the accuracy of street crime risk assessment and reduces the false alarm rate.Finally,we summarized the main research results and innovations of this thesis,and points out the future research directions,mainly including:1)Exploring the effect of the spatial distribution structure of multielements of urban built environment on the heterogeneity of street crime risk distribution;2)Investigate the comprehensive causal explanation mechanism of human travel,geographical environment and social economy on the risk distribution of street crime;3)Study the intelligent early warning method of dynamic risk of street crime guided by prior mechanism.