The　effect　of　the　semantic　segmentation　methods　has　been　significantly　improved　by　introducing　deep　convolutional　neural　networks　(CNNs)　in　recent　years.　But,　these　approaches　need　quantities　of　data　with　pixel-level　labels　and　have　poor　generalizability.　Few-shot　segmentation　(FSS)　is　subsequently　proposed,　which　can　complete　the　segmentation　of　unseen　classes　with　only　a　small　amount　of　annotated　data.　However,　existing　approaches　have　the　following　two　problems.　First,　they　ignored　the　connection　between　feature　channels　of　support　features　and　query　features.　Second,　they　did　not　fully　mine　the　information　in　the　support　set.　These　two　problems　harm　the　segmentation　performance.　In　this　study,　we　design　a　new　FSS　approach　to　alleviate　the　above　two　problems.　Our　model　consists　of　two　parts:　the　channel　attention　module　is　designed　to　fully　mine　the　connection　of　the　feature　channels　of　support　and　query　set,　and　the　supervision　augmentation　module　is　designed　to　fully　utilize　the　information　in　the　support　set　and　strengthen　the　supervision　of　the　model.　Extensive　experiments　verify　that　our　model　achieves　0.9%　and　1.3%　mIoU　(mean　intersection　over　union)　improvements　over　the　baseline　on　the　PASCAL-5i　dataset　for　1-shot　and　5-shot　tasks,　respectively,　and　our　approach　is　more　lightweight　than　other　state-of-the-art　(SOTA)　approaches.　©　2023　IEEE.