Abstract:Insulator defect detection in distribution networks is a crucial task for ensuring the safe operation of power systems. To address the challenges of complex small-object features and small scales in insulator defect detection，improvements are made based on the YOLOv9 algorithm that proposed to optimize the performance in both network structure and loss function. Firstly，to overcome the limitations of traditional attention mechanisms，a quad attention module （QAM）is proposed，adding a spatial attention branch to the existing triple attention module （TAM）. This enhancement retains the model’s ability to capture cross-dimensional interactions while strengthening its focus on the specific spatial location of insulator defects，thereby improving its ability to extract features from small objects. Secondly，the Power-IoU loss function is introduced to accelerate model convergence. Experimental results on public datasets show that the improved model achieves 82.8% M_map@0.5 with a detection speed of 81 FPS on the insulator defect detection task，representing a 3.6% improvement over the original YOLOv9 model.