Despite numerous improved techniques being widely applied in industrial defect detection, current methods still face three major challenges. First, complex background interference makes accurately distinguishing real defects from normal backgrounds particularly difficult. Second, precise detection of tiny defects in actual production environments is quite challenging. Third, efficient detection of various industrial defect types also presents challenges. Addressing these challenges, this paper proposes an automatic industrial defect detection method called ESN-DETR based on RT-DETR. We designed a novel backbone network EAAB that integrates edge information, and constructed a SEMFPN feature pyramid that introduces P2 shallow features for bidirectional transmission. Additionally, we developed the NAC-Attention mechanism to enhance the model's capability to capture complex channel relationships. We conducted extensive experiments on NEU-DET, GC10-DET, and fabric defect datasets. Experimental results show improvements of varying degrees in \({\rm{AP}}_{\rm{s}}^{50-95}\) , \({\rm{AP}}_{\rm{m}}^{50-95}\) , \({\rm{AP}}_{\rm{l}}^{50-95}\) , \({\rm{AP}}^{50}\) , Loc, Miss, and Gflops metrics, with overall \({\rm{AP}}^{50}\) indicators increasing by 4%, 5.8%, and 0.8% respectively. The computational load (Gflops) was also reduced by approximately 10%.