The present data is shared in order to reproduce a study aimed to automatically extract and identify targets (gas seeps, fish schools and noise) from the water column using machine learning techniques. The dataset is composed of two surveys conducted over a square area of 1X1 Km wide centered on a four-leg gas platform located in the Central Adriatic Sea at a distance of 65 km from Ancona (Italy). Surveys were conducted in calm sea conditions the 11-12-2018 and 24-07-2019. The instrument used is a hull-mounted Multi-Beam Echosounder Simrad Knogsberg EM2040CD, acquisition frequency is 300 kHz, the opening angles were 80°/10° (port, starboard) for one head and 10°/80° (port, starboard) for the other head maintaining 10° of overlap between the two swaths. The seabed depth is about 85 meters, the survey was conducted following 10 parallel lines N-S oriented with a distance of 150 meters one from the other in order to assure at least a superposition of 50% between one row and the other. A Trimble SPS855 GNSS Receiver with Omnistar HP/XP/VBS subscription was used for ship positioning, while a Kongsberg motion sensor MRU 5 and an Anschutz Compass Standard 20 Compact Type 110-222 NG001 gyro corrected pitch, roll, heave and yaw movements. Sound velocity profiles were systematically collected with an AML Oceanographic Smart SV&P sensor before the surveys were started (and repeated when needed), while a Valeport MiniSVS probe was mounted close to the transducers to continuously measure the sound velocity for the beamforming.