The growth of human activities along coastal regions has prompted the creation of Marine Protected Areas (MPAs), where stringent regulations or outright bans on human activities are imposed. Yet, without vigilant monitoring, these areas can become magnets for unlawful behavior. Monitoring human activity at sea is usually facilitated through the Automatic Identification System (AIS), a vital tool for real-time vessel identification and collision avoidance. AIS data provides crucial information about vessel positions (e.g., longitude, latitude, speed) and voyage specifics, including vessel types, but it can be disconnected at any time and it is not mandatory for certain vessel types. Therefore, monitoring these MPAs presents several challenges, particularly due to the limitations of visual surveillance. Because sound travels further and better than light underwater, passive acoustic data is a major candidate to monitor such activities. For this reason, in this study, we first create a comprehensive database using passive acoustic hydrophone recordings collected over 100 days from two Belgian North Sea stations and AIS data, forming the cornerstone for identifying and classifying marine vessels based on their acoustic signatures in shallow waters. Then, by employing advanced machine learning techniques adept in object detection and classification, this endeavor aims to establish the groundwork for robust underwater acoustic surveillance systems. Thus, as a solution for safeguarding MPAs and other protected areas such as wind farms, this initiative advocates for the implementation of an underwater acoustic monitoring system to discern vessels through their unique sound signatures.