Abstract: Vacuum degree is a critical factor influencing the insulation strength and interrupting performance of vacuum circuit breakers. To address the limitations of traditional vacuum degree detection techniques for vacuum interrupters, such as low accuracy, stringent equipment requirements, and complex operation, this paper proposes a vacuum degree detection method for vacuum interrupters based on high-frequency high-voltage power supply excitation and a dual-channel convolutional neural network (CNN). Experimental studies are conducted within the engineering-relevant vacuum degradation range of 10⁻⁴ to 10⁻² Pa. Based on the experimental results, a dual-channel CNN is further constructed to extract and fuse time-domain and frequency-domain features, achieving intelligent identification of five vacuum degree states. The model achieves a validation accuracy of 91.62% without obvious overfitting, exhibiting good generalization performance. The proposed method provides a new technical approach for on-site non-destructive testing of vacuum interrupters.