An adaptive genetic algorithm optimizing neural network (AGA-BP) applied in the gun-launched ground vibratiuncle sensor recognition was introduced. First, as for the BP neural network may not converge to the global minima, an optimization algorithm formed by combination of adaptive genetic algorithm with the BP neural network was proposed and the network was trained by the back-propagation method. Then the simulation was conducted for time-frequency analysis on sampling signals of tracked vehicle and wheeled vehicle, the eigenvalue was extracted from sampled signals by wavelet transform. The sample training and recognition were performed by optimized algorithm and traditional algorithm, the comparison shows recognition error can be reduced by the method.