基于少样本学习和知识迁移的继电保护压板状态识别

Protection Plate Status Recognition Based on Few-Shot Learning and Knowledge Transfer

针对变电站中保护压板开关状态自动识别问题,提出了一种基于少样本学习和知识迁移的压板开关状态识别模型,使用残差网络提取图像特征,基于度量方法计算查询图像与支持图像之间的相似度,在此基础上使用KNN实现压板开关状态的分类识别。将残差网络在公用数据集上预训练的模型直接迁移到基于少样本学习的压板开关状态的识别任务,并研究了KNN算法中不同的最近邻个数对压板开关状态分类结果的影响。所提方法可在图像样本少的情况下实现压板开关状态的识别。实验结果表明,在支持图像的样本数为30时,图像识别精度达到99.49%。相比于其他大样本的分类方法,所提出的利用少量样本的分类方法能够实现令人满意的分类效果,提高了图像分类的效率。

Aiming at the issue of automatic recognition of the switch status of the protection plate in the substation,a recognition model of the protection plate switch state based on few-shot learning and knowledge transfer is proposed.The residual network is used to ex-tract image features,and the measurement method is used to calculate the difference between the query image and the support image.Based on the similarity,KNN is used to realize the classification and recognition of the switch state of the protection plate.The residual network pre-trained model on the public data set is direct transferred to the recognition task of the protection plate switch state based on few-shot learning,and the influence of different number of nearest neighbors in KNN algorithm on the classification results of the switch state of the protection plate is studied.The proposed method can realize the recognition of the switch state of the protection plate when there are few image samples.The experimental results show that when the number of supporting samples is 30,the image recognition ac-curacy reaches 99.49%.Compared with other large sample classification methods,the proposed classification method using a small num-ber of samples can achieve satisfactory classification results and improve the efficiency of image classification.