Abstract: Insulator defects with different degrees have similar features and less pixel information, resulting in poor detection effect, therefore, an insulator defect degree detection network based on multi-scale feature fusion(MFFD³Net) is proposed. The network uses reconstructed ResNeSt50 to improve the feature extraction ability in insulator defect dataset. A multi-scale feature fusion module based on deconvolution is designed, which enriches the expression ability of different size feature maps and improves the detection performance of different scale targets. At the same time, the receptive field block(RFB) is added after the shallow feature maps of the input detection module to ensure more insulator defect information to enter the effective receptive field, which has an impact on the final feature map and improves the detection accuracy of insulator defects in different degrees. The mAP of MFFD³Net on insulator defect degree dataset reaches 85.02%, the detection accuracy of small targets such as slight breakage and slight flashover is 78.37% and 79.98%, which can complete the identification and location of insulator defects in different degrees. Thus, the MFFD³Net proposed in this paper is of great significance for improving the fault warning of power system and ensuring the safe and stable operation of power grid.