Abstract: To solve the low recognition rate problem caused by the incompleteness of image information and noise interference caused by the traditional disconnector opening and closing state recognition method in the single image mode, a fusion method based on a Multi-scale Densely Connected Pyramid Attention Network (MSDPAN) was proposed that fused the infrared and visible images of the high-voltage disconnector. A Faster Region-based Convolutional Neural Network (Faster R-CNN) algorithm was used to recognize the fused images to improve the accurate identification ability of the opening and closing states of the high-voltage disconnector. The experimental results showed that the fusion strategy based on the MSDPAN retained a large amount of image information and was superior to several other common fusion strategies in terms of various evaluation indicators. The accuracy rate of Faster R-CNN reached 95.24% that was 6.15 percentage points higher than that of the CNN. Moreover, the average accuracy of the fusion image for the state recognition of the high-voltage disconnector reached 92.67% that was 7.67 percentage points higher than that of the infrared image and 3.33 percentage points higher than that of the visible image.