Abstract: In view of the low efficiency of manual inspection of the status of the transformer substation, and the poor effect of the image recognition of the existing torsional Angle clamp, a new method of the status recognition of the clamp based on the semantic segmentation model of VGG16-Unet was proposed.Firstly, the network model structure of VGG16-Unet is designed, which includes the trunk feature extraction network part, the enhanced feature extraction network part and the prediction network part, and captures the deep semantic features and shallow detail features of the image during the down-sampling and up-sampling of the network.Secondly, the Dice loss function of network model is defined and four performance evaluation functions are analyzed to detect the recognition effect of platen.In the final experiment on a dataset of 1000 pressure plates, the deep learning method achieved an accuracy of 98.6%,a recall rate of 92.2%,a comprehensive index of 95.3%,and an average intersection to union ratio of 92.4%.Compared with the existing mainstream pressure plate state recognition methods, the results show that the proposed method has better recognition performance.