To estimate the location of a low-frequency target in the complex environment of shallow seas, both real-time properties and accuracy must be considered. In this paper, a multichannel information fusion method is proposed for estimating the location of a target based on a single vector hydrophone. First, a window signal fusion algorithm is proposed, which combines the EM algorithm to achieve adaptive signal extraction. The RNN used later realized the self-localization of the sound source. The results show that the fixed-dynamic window based on the steepest rising segment of Shannon entropy can divide a very short signal sample into a sufficient number of signal sets and the signal self-replenishment based on the EM algorithm can further strengthen the characteristics of different signal segments on the basis of replenishing the signal. Compared with other networks, the RNN has the highest accuracy and stability for this type of location estimation of an acoustic source based on the time-domain signal. The experimental results indicate that in the shallow sea environment, this model only depends on a single vector hydrophone to collect signals and rapidly locate the acoustic source, and the error radius is controlled within 1.5 m.