Abstract: Object detection technology based on deep learning has been widely used in the insulator defect detection. However, the existing object detection algorithms are mainly based on the abundant defect samples to train the network models, which is unable to identify the defects with few samples accurately. To solve the problem of insufficient defect samples in the insulator defect detection, this paper proposes a novel few-shot insulator defect detection based on the deep information of local features. Firstly, the insulator strings are extracted using the oriented R-CNN (oriented region-based convolutional neural network). Next, the insulator string features are divided into sub-blocks and the local features are employed to realize the few-shot defect detection based on a deep EMD (earth mover's distance) network. The experimental results show that the proposed method with 2 training samples can achieve the same results as those of the existing object detection method with 200 training samples for the self-explosion defect detection of glass insulators. The mAP (mean average precision) of insulator self-explosion detection with 10 training samples is up to 0.65. The proposed few-shot defect detection method provides a new solution and an implementation method for the intelligent defect detection of the power equipment with few defect samples.