The threat of terrorism has been increasing in various venues, including in the air, on land, on the water, and underwater. Cameras and radar that use light and radio waves are not sufficiently effective underwater. So underwater regions have become especially prone to security gaps. To secure such gaps, we have been developing an underwater monitoring system that can always automatically detect intruders from underwater. We conducted a demonstration experiment involving a diver and ROV detection near Numazu Bay and Tokyo Bay to confirm the performance of the underwater monitoring system. In extremely shallow waters, it is difficult to detect moving objects by underwater acoustic methods, and false alarms are frequently generated with conventional detection methods. In this paper, we used deep learning techniques to improve target detection performance by suppressing noise and reverberation. We confirmed that our approach worked to suppress noise and reverberation significantly.