The　paper　presents　a　structure-preserving　domain　adaptation　network,　termed　SDANet.　To　address　the　problem　of　edge　blurring　or　incompleteness　in　the　pencil　sketches　generated　by　existing　methods,　we　designed　a　structure-preserving　module　to　extract　the　edges　of　both　the　real-world　photos　and　the　generated　real-world　photos,　then　constrained　them　with　a　structure-preserving　loss,　in　order　to　produce　clearer　edges　in　the　generated　pencil　sketches.　In　addition,　we　constructed　a　real-world　photo　and　pencil　sketch　dataset　for　training　the　pencil　sketches　generation　network,　named　RW-PS,　which　comprises　an　extensive　collection　of　high-quality　pencil　sketches.　Experimental　results　demonstrate　that　our　method　generates　pencil　sketches　with　a　lowest　FID　score　compared　to　existing　SOTA　methods　on　the　RW-PS　dataset.　Moreover,　in　user　surveys,　the　pencil　sketches　generated　by　our　proposed　method　received　the　highest　number　of　votes.　©　2023　IEEE.