Addressing　high　impedance　faults　(HIF)　in　resonant　distribution　networks　remains　a　formidable　challenge.　This　paper　introduces　a　multitask　learning-based　approach　integrated　with　a　multitask　fault　detection　network　(MTFD-Net),　employing　three　task-specific　heads—classification,　segmentation,　and　regression—to　enable　precise　fault　detection.　MTFD-Net　utilizes　zero-sequence　voltage　data　and　a　sliding　time　window　to　perform　initial　coarse　classification,　which　allows　the　classification　head　to　determine　whether　a　permanent　HIF　has　occurred.　Upon　detection,　MTFD-Net　proceeds　to　pinpoint　potential　fault　moments　through　the　outputs　of　the　segmentation　head.　The　regression　head　further　refines　these　moments　by　predicting　a　reference　moment　and　calculating　the　distance　to　each　potential　fault,　effectively　isolating　the　exact　fault　moment.　An　industrial　prototype　was　developed　and　rigorously　tested　on　a　10　kV　system,　where　MTFD-Net　demonstrated　superior　performance,　achieving　an　accuracy　of　0.976,　an　intersection　over　union　of　0.984,　and　an　absolute　detection　deviation　of　5.20　ms.　Operating　efficiently　with　inference　times　ranging　from　9.69　to　15.45　miliseconds　on　a　Raspberry　Pi　4B,　MTFD-Net　surpasses　existing　methods　in　accuracy,　F1-score,　sensitivity,　specificity,　and　detection　accuracy,　providing　a　robust　solution　for　HIF　detection　in　resonant　distribution　networks.　©　2025　Elsevier　Ltd