Insulators which are widely used in all aspects of the power system play an important role in ensuring the safe and stable operation of the power grid. Existing methods are only able to identify obvious defects such as self-explosion missing and foreign objects, and cannot deal with local damage, cracks and other situations. In response to the above problems, an insulator defect detection method based on enhanced feature pyramid and deformable convolution is proposed. On the basis of the original high and low feature fusion, an enhanced bottom-up path is added, which improves the information transfer between high and low feature maps, and realizes the effective extraction of local defect features. The introduction of deformable convolution, adaptively changes local sampling points, reduces the impact of background interference, and further improves the applicability of the model. Comparative experiments using insulator images collect in multiple scenes show that the proposed method achieves greater detection accuracy improvements on different basic networks, and can be widely used in various insulator application scenarios such as substations and high-voltage transmission lines.