In order to build a green and low-carbon power grid, the State Grid takes the SF6 gas content in the low-pressure gas filling cabinet in the small gas chamber as an important detection index. The existing SF6 gas on-site detection technology consumes a lot of gas, and the gas density of the gas chamber decreases significantly after repeated detection in the low-pressure gas filling cabinet, which affects the insulation performance. In order to solve the above problems, a quantitative gas extraction detection technology for tracing SF6 gas in environmentally friendly inflatable cabinets based on a neural network model was proposed. Through theoretical analysis and PCCs calculation, it is verified that the output electrical signal of the pyroelectric probe at different times in the quantitative gas extraction mode has a linear relationship with the SF6 gas concentration. The experimental results show that the two neural network models can accurately output SF6 gas concentration in the quantitative gas intake mode, the main performance indicators of the GA-BP neural network model are good, and the maximum relative error is -2.7%, which is significantly better than the BP neural network model.