For multi-sensor network systems with uncertain noise variances, traditional fusion estimation algorithms need to optimize multi-dimensional nonlinear cost functions, resulting in greater computational complexity. This paper proposes a fast sequential covariance cross-fusion adaptive unscented Kalman filter algorithm (SCI-AUKF), which mainly solves the optimization problem of multiple one-dimensional nonlinear cost functions. It is a recursive two-sensor filter. Its accuracy is higher than that of local estimators, and it will effectively solve the problem of state estimation under uncertain noise variance. Through simulation, the filtering method is compared with the other filtering methods, and its accuracy is significantly improved. An application example of the algorithm in radar nets is given, which shows the superiority and effectiveness of the SCI-AUKF filtering method.