Detection and Classification of Subsea Objects in Forward-Looking Sonar and Electro-Optical Sensors for ROV Autonomy

SeeByte and Greensea Systems, Inc. (Greensea) have teamed up with the sponsorship of the Defense Innovation Unit (DIU) and PMS 408 to advance Automatic Target Recognition (ATR) of Mine Like Objects (MLOs) for the purpose of developing an autonomous capability for Remotely Operated Vehicles (ROVs) used in maritime Explosive Ordnance Disposal operations. Advancements in autonomy requires advancements in the sensor processing to detect the surrounding environment and identify objects of interest. In this case the sensors chosen were a dual frequency Forward Looking Sonar and a Electro-Optical (EO) stereo camera system. The objects of interest are MLOs that need to be located, identified, and inspected by an autonomous submersible robot. SeeByte used Deep Learning Neural Network (DNN) on both of these sensor feeds yielding a very robust detection and classification system. The output of that detection and classification is provided to Greensea's vehicle control and autonomy software as candidate targets for mapping and prosecution.