When　EEG　signals　are　used　to　assess　the　level　of　student　engagement　in　online　teaching　tasks,　they　are　often　interfered　by　noise.　It　is　a　challenge　to　effectively　remove　these　noises.　Currently,　deep　learning　methods　have　been　applied　to　the　field　of　EEG　denoising.　However,　existing　methods　still　have　some　problems.　The　denoised　signals　still　have　obvious　noise　residue,　or　the　original　EEG　signals　are　damaged,　and　the　model　fitting　speed　is　too　slow.　Image　dehazing,　as　a　typical　denoising　task　in　the　field　of　image　enhancement,　has　achieved　great　success　in　recent　years.　Therefore,　inspired　by　advanced　models　in　this　field,　we　introduce　CNN　into　EEG　denoising　tasks.　In　this　paper,　we　take　GCANet,　an　excellent　image　enhancement　model,　as　an　example.　The　dilated　convolutions　and　gate　fusion　subnetworks　included　in　GCANet　enable　more　efficient　EEG　signal　denoising.　The　results　demonstrate　that　the　proposed　model　effectively　reduces　noise　while　preserving　essential　features.　Furthermore,　in　comparison　to　other　state-of-the-art　models,　our　proposed　model　exhibits　enhanced　robustness　and　faster　convergence,　as　evidenced　by　achieving　lower　loss　values　after　five　epochs.　Its　good　performance　provides　a　new　development　idea　for　the　field　of　EEG　denoising.　©　2023,　Taiwan　Ubiquitous　Information　CO　LTD.　All　rights　reserved.