With　the　rapid　advancement　of　edge　computing　technology　and　the　widespread　application　of　artificial　intelligence,　the　deployment　of　neural　network　inference　at　the　edge　has　garnered　increasing　significance.　However,　constrained　by　limitations　in　computational　resources　and　security　considerations,　effectively　verifying　the　correctness　of　neural　network　inference　at　the　edge　poses　a　formidable　challenge.　To　address　this　challenge,　this　paper　proposes　a　neural　network　inference　verification　framework　based　on　the　generalized　GKR　protocol,　specifically　tailored　for　edge　deployment.　Leveraging　the　bidirectional　efficiency　inherent　in　the　generalized　GKR　protocol,　this　framework　enables　rapid　and　precise　validation　of　neural　network　inference,　thereby　enhancing　the　reliability　and　security　of　edge-based　neural　network　inference.　Additionally,　this　paper　tackles　challenges　encountered　in　designing　the　verification　framework,　such　as　handling　model　parameters　and　transforming　non-linear　functions,　utilizing　pertinent　techniques.　Finally,　the　effectiveness　and　performance　of　the　framework　are　validated　and　analyzed　through　experimental　verification.