DVGN: A Differential Value Guidance Network for Underwater Objects Detection

Although considerable progress has been made in optical underwater objects detection (UOD) in recent years, the task still faces great challenges due to the complexity of the underwater environment and the mimetic behaviour of underwater objects. Existing deep learning-based models usually do not consider the enhancement of underwater object features, resulting in inefficient UOD. In order to solve the above problems, this paper proposes a plug-and-play Differential Value Guidance Network (DVGN) for arbitrarily embedding into an existing model to improve the recognition performance of the model. The salient features of this paper are: 1) We propose a plug-and-play efficient feature fusion method. It can focus on features at different scales and enhance the recognition capability of objects at different scales underwater. 2) Through a large number of experiments, it is proved that it can be embedded into different object detection algorithms to effectively improve the object detection capability with the smallest possible increase model parameters.