EUDet: Power Equipment Defect Detector Driven by Elastic Unit

Equipment security in power business scenarios plays an important role in the security of the power system. This paper proposes a de-long-tail data enhancement strategy to address the problems of long-tail distribution of defect data, low power defect detection accuracy, and poor model detection of multi-scale targets in power business scenarios. Through two-stage adaptive data enhancement and random resampling method to alleviate the long-tail distribution of the dataset. Secondly, we proposed the Elastic Feature Extraction Unit (EFEU), which is based on selective kernel convolution and fully fuses the extracted features to improve the multi-scale target detection capability of the model and further improve the accuracy of the model. Finally, we built an efficient detector specifically for power equipment defect detection based on EFEU and conducted sufficient experiments on the substation data set, with mAP reaching 85.3%.