Under high voltage environment, the leakage process of SF6 gas in key parts of circuit breaker joints and seals becomes more complicated and rapid due to discharge, overheating and the presence of water vapor. Under such conditions, SF6 gas will rapidly decompose and generate harmful gases under the dual action of high electric field strength and electromagnetic interference, which not only aggravates the complexity of the leakage process, but also increases its difficult to directly observe the characteristics. Therefore, a study on the rapid early warning algorithm for SF6 circuit breaker gas leakage under high voltage decomposition is proposed. The potential area of SF6 gas leakage is extracted using the pixel-level adaptive segmentation algorithm and the adaptive threshold inter-frame difference method. The problems of internal cavities, incomplete edges, and incoherent regions in the potential area are processed through morphological opening and closing operations and regional connectivity methods. Then, based on the area growth characteristics of gas leakage and diffusion, the leakage gas is distinguished from other moving objects to eliminate false alarms, false positives, and other situations. The average differential pixel points are quantified as the degree of leakage, and the rapid early warning of SF6 circuit breaker gas leakage is performed according to the alarm threshold. Experimental results show that the proposed method has high throughput rate, target contour integrity, and intersection-over-union in the detection of potential areas of SF6 circuit breaker gas leakage, and a high AUC value in early warning.