Boundary　information　of　agricultural　fields　is　essential　to　many　agricultural　applications,　particularly　at　the　field　level.　This　paper　investigates　the　use　of　high-resolution　remote　sensing　images　to　delineate　the　boundaries　of　agricultural　fields.　We　consider　the　delineation　task　a　multi-Task　semantic　segmentation　problem　and　use　a　recent　deep　neural　network,　i.e.,　Psi-Net,　to　do　the　semantic　segmentation.　The　structure　of　a　Psi-Net　consists　of　an　encoder　and　three　decoders.　The　decoders　learn　three　parallel　tasks,　corresponding　to　a　primary　task　(i.e.,　mask　prediction)　and　two　additional　tasks　(i.e.,　contour　detection　and　distance　map　estimation).　The　additional　tasks　are　used　to　regularize　the　mask　prediction　to　produce　a　refined　mask　with　smooth　boundaries.　We　conducted　experiments　on　a　GF1　PMS　satellite　image　(2m)　acquired　in　the　21st　regiment　of　the　2nd　agricultural　division　of　Xinjiang.　To　evaluate　the　effectiveness　of　the　proposed　method,　we　compared　it　with　existing　single　task　semantic　segmentation　using　UNet.　Our　results　show　that　the　proposed　method　using　Psi-Net　performed　better　than　the　existing　method　from　the　perspective　of　geometric　and　attribute　accuracies.　We　conclude　that　the　proposed　Psi-Net　method　has　a　high　potential　for　extracting　field　boundaries　from　high-resolution　remote　sensing　images.　©　2022　IEEE.