Imaging sonar is commonly used in underwater detection tasks, as it can provide acoustic imaging of underwater environments in low or zero visibility conditions where optical sensors fail. The forward-looking sonar (FLS) is commonly used for monitoring tasks due to its capability of providing an acoustic 2D image. The interpretability of such images is yet an issue, however, as it is not always easy for the non-trained eye to comprehend the scenery or detect features. In DFKI’s DeeperSense project, an imaging FLS aboard an autonomous underwater vehicle (AUV) is used to monitor technical divers performing various underwater tasks. The collected datasets contain considerable diversity in diver positions while keeping the diver in a direct field of view. This thesis aims to develop an object detector to identify underwater divers and determine their relative position to the AUV, utilizing deep learning techniques, particularly convolutional neural networks (CNN). State-of-the-art CNN architectures have shown better performance in computation, frames per second, and model complexity when it comes to object detection with sonar images.