In order to monitor the closing state of the electric isolation switch, the image recognition method is used to monitor whether the isolation switch is closed to ensure the safety of the maintenance personnel. A fusion NSCT and two-dimensional maximum entropy segmentation method are proposed to segment the image, and the region of interest (gate knife) is extracted. Then, the feature extraction of the blade closing and closing is performed by the pixel integral projection method, and the extracted image will be extracted. The eigenvalues are imported into the BP neural network for training, and a classifier that can automatically identify the position of the knife is obtained. The collected pictures are imported into the BP neural network classifier for identification experiments. Experiments show that the image after processing has strong anti-noise ability, and the trained BP neural network has a high recognition rate of the closing state of the knives, which is more than 95%. This work is supported by National Natural Science Foundation of China (No. 51767006), the Key R & D Projects of Jiangxi Province (No. 20181BBE50019), Cultivate Plan for Application Research of Jiangxi Province (No. 20181BBE58015), and Science and Technology Project of Jiangxi Provincial Department of Education (No. GJJ170378).