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.　©　2024,　Editorial　office　of　PROGRESS　IN　GEOGRAPHY.　All　rights　reserved.