Power Equipment Object Detection Based On Incomplete Supervision Learning

It is commonly believed that fast and accurate object detection is the prerequisite for power equipment automatic fault diagnosis.In recent years,with the deepening of the research on a series of object detection algorithms based on deep learning,the relevant methods have also become the mainstream in the field of power equipment object detection.However,in order to obtain high detection accuracy to support practical applications,the existing methods often need a large number of high-quality labeled data to meet the training needs of complex models.Researchers need to spend a lot of time and manpower to label these images.While insulators detection requires a huge number of labeled images to guarantee its performance,placing bounding boxes for every object in each image is time-consuming and requires professional knowledge.To alleviate this problem,a two-stage semi-supervised learning framework for power line insulators detection based on consistency regularization and data augmentation(CASD)is proposed.Firstly,train a teacher model on labeled images,then generate pseudo labels for unlabeled images using the trained model,and the confidence-based thresholding is introduced to control the quality of pseudo labels.In the second stage,strong data augmentation is applied to unlabeled images and both the labeled and unlabeled images are used to train the CASD.Consistency regularization is introduced in the calculation of unsupervised loss to improve the accuracy and robustness of the model.Experiments show that the proposed method performs better than all the three baseline methods,and the detection accuracy of CASD is far better than the supervised learning method in the case of less available images.The comparison results further prove the efficacy of the proposed method.The fault diagnosis of power equipment often depends on multimodal images,so it also puts forward requirements for the crossmodal domain adaptability of the object detection model.This thesis proposed an unsupervised domain adaptative Faster R-CNN for surge arrester detection,using the labeled visible images of surge arrester as the source domain data and using the unlabeled infrared images of surge arrester as the target domain data.Firstly,a theoretical analysis for cross-domain object detection from a probabilistic perspective is carried out,put forward a solution to align the data distribution of source and target domain from image level and instance level.Integrate the proposed image-level adaptation component and the instance-level adaptation component into the Faster R-CNN model,and implement them by learning a domain classifier in adversarial training manner.The proposed approach is evaluated using multiple datasets.Compared with the original model,the unsupervised domain adaptive Faster R-CNN proposed in this thesis has significantly improved the detection accuracy on the infrared image test set,which proves the effect of this method in improving the cross modal generalization ability of the model.