In the context of the mine warfare modernization plan, the French Ministry of Defense is studying the possibility of close object inspection by an autonomous underwater vehicle. To that end, the vehicle is equipped with a multibeam echosounder (MBES), and travels to within 10 meters of the object. With each narrow beam formed by the MBES, a sounding is measured by a detection algorithm that estimates the two-way propagation time between the sonar and the seabed. Unfortunately, manufacturer algorithms are designed for bathymetry estimation and suffer from two drawbacks for object recognition: first, multiple detections per beam are seldom available; second, too many false alarms arise in case of strong specular reflection. In order to achieve better recognition of small objects, in this paper we propose processing the water column data where the 3D shape of the object is visible. This huge volume of data needs to be reduced, however. Data are first enhanced by the use of the bitonic filter, which combines non-linear morphological and linear operators. Then, hysteresis thresholding is applied, allowing multiple detections, as well as false alarm mitigation. This technique is assessed on various data sets collected by the French Defense Procurement Agency (Direction Générale de l'Armement, Naval Techniques section) with three different MBES: R2Sonic2022 (700kHz), EM2040 (400kHz) and MB2250 (2.25MHz). Results show that the set of detected points allows us to complete the final step of recognition based on 3D shape matching.