To solve the problem that the algorithm effect of disconnecting switch state recognition is limited by training samples

an augmentation method of disconnecting switch images combined with three-dimensional spatial information is proposed. Firstly

through the two-dimensional images

along with the prior knowledge about the three-dimensional shapes of the disconnecting switches

the three-dimensional coordinates of the disconnecting switch key points are solved. Then

the disconnecting switches are rotated and transformed for several times in a three-dimensional space to simulate the change of shooting angles. Finally

the disconnecting switches in a three-dimensional space are projected to a two-dimensional plane by perspective projection transformation

in order to generate new images and augment training samples. The results of the case study show that

compared with the traditional image augmentation methods

such as translation

scaling

rotation

changing aspect ratios and changing brightness

the proposed image augmentation method can improve the position and state recognition effect of disconnecting switches more significantly.