Spatial-temporal Analysis Of Urban Crimes And Scene Risk Forecast

The rise of urban crime rate poses a serious threat to the construction of "safe city" and the harmonious development of society.Based on the spatial-temporal quantitative analysis method and deep learning technology,taking Chicago as an example,this paper analyzes the spatial-temporal differentiation mode of robbery crime and its causes,builds a crime risk prediction model based on high-resolution remote sensing images,and introduces convolution attention mechanism to automatically capture the urban scene features that have an important indication of crime risk,which verifies the effectiveness of remote sensing images on crime risk prediction the feasibility and effectiveness of crime risk prediction.The research results show that:(1)there are significant differences in the day,week,month,season and month distribution of robbery,and the crime time has a stable hot spot;(2)Robbery cases are clustered in spatial distribution,and there are stable spatial hot spots;(3)It is found that the spatial distribution of robbery cases has positive correlation with the proportion of African American population,unemployment rate,road density and the number of business POI,and has negative correlation with the annual family income,population density,and subway station.At the same time,the results of geographical weighted regression analysis show that the spatial distribution of each influencing factor is different;(4)It is verified that the features of remote sensing image are related to the crime risk.At the same time,the network with convolution attention module of Res Net＿50 can get 75.2% of the accuracy rate in the crime risk prediction task of urban scene,and the accuracy rate is 0.27 higher than Res Net＿50.In addition,this paper explores the urban scene features that contribute to the high crime risk by class activation mapping to visualize the image area of the network "attention".The gray,black,white roofs and dense roadside trees in the high-density low-rise residential area scene and the parking lot features in the commercial area scene are obviously related to high-risk robbery.In this paper,the crime spatial-temporal analysis and crime risk prediction model based on urban spatial-temporal big data enrich and develop the research perspective and technical methods of urban crime geography,provide new theoretical basis and tools for exploring the occurrence of crime,and provide technical framework and reference for low-cost,large-scale and high-precision urban crime inducement identification and crime risk prediction scientific basis.