Researching on insulator defect recognition based on context cluster CenterNet+.

The existing UAV inspection images are faced with many challenges for insulator defect recognition. A new multi-resolution Context Cluster CenterNet++ model is proposed. First, this paper proposes the Context Cluster method to solve the problem of low recognition accuracy caused by non-uniform distribution of targets. The cluster region is used to identify and predict the location of the target, and the improved loss function is used to modify the cluster center. Secondly, this paper uses deformable convolution operator (DCNv2) combined with path aggregation network (PAN) to carry out deformable convolution operation on the image, and accurately predicts the regression box and key point triplet (KP), so as to improve the accurate positioning of the target position of any shape and any scale. The sensitivity of the model to target scale change and deformation is reduced, and the recognition accuracy of the model is improved. Then, Bhattacharyya distance is used to calculate the triplet prediction loss of key points and the target center point offset loss, so as to significantly improve the positioning accuracy of the same target in different frames. Finally, experiments are carried out on the MS-COCO dataset and the National Grid standardized UAV inspection insulator image dataset. Our code is at https://github.com/mengbonannan88/CC-CenterNet .