Urban　parks　have　been　found　to　provide　mental　health　benefits.　Some　empirical　studies　have　tested　natural　features　and　perceptual　measures　respectively,　announcing　their　contribution　to　psychological　restoration.　However,　inconsistent　findings　were　occasionally　reported,　whereas　few　attempts　have　been　made　to　combine　both　observed　and　perceptual　factors　for　validation.　Little　is　known　about　the　variation　of　restorative　drivers　and　their　spatial　patterns.　To　address　these　problems,　this　study　combined　public　participation　geographic　information　system　(PPGIS)　and　deep　learning　method　to　capture　visual　qualities　of　landscape　features　along　with　several　important　perceptual　measures.　A　typical　urban　park　in　Wuhan,　China,　was　selected　for　a　pilot　study,　and　1560　crowdsourced　on-site　images　were　collected,　with　thematic　and　geographic　information　being　integrated.　A　series　of　statistical　models,　e.g.,　OLS,　QRM,　and　MGWR,　were　employed　successively　for　validation.　The　results　showed　that　landscape　preference,　place　attachment,　greenery　and　water　were　validated　as　the　global　explanatory　factors　to　estimate　the　conditional　mean　of　psychological　restoration.　The　variation　of　influential　effects　of　these　factors　were　detected　at　different　restoration　levels.　There　exist　spatial　heterogeneity　for　these　influential　factors　on　restorative　effects.　Findings　provided　new　knowledge　on　a　deeper　understanding　of　the　subtlety　of　restoration　drivers　and　their　spatial　patterns.　The　findings　offered　useful　insights　and　guidance　for　urban　planners　in　creating　high-quality　green　parks　with　restorative　values.　©　2024　Elsevier　Ltd