More and more underwater robots are deployed to investigate marine biodiversity autonomously, and tools are needed by underwater robots to discover and acknowledge marine life. This paper has proposed a convolutional neural network-based method for intelligent fish detection and recognition with a dataset used for training and testing generated and augmented from an open-source Fish Database regarding 6 different types. Firstly, to improve image quality, a hybrid image enhancement algorithm is used to preprocess underwater images with a weighted fusion strategy of multiple traditional methodologies and comparisons have been made to prove the effectiveness according to various indexes. Secondly, to increase detection and recognition accuracy, different attention modules are integrated into the YOLOv5m network structure and the convolutional block attention module(CBAM) has outperformed other modules in recall rate and mAP while maintaining the capability of real-time processing. Lastly, to meet real-time requirements, lightweight adjustments have been made to CBAM-YOLOv5m with the GSConv module and C3Ghost module and a nearly 25% reduction in network parameters and a 20% reduction in computational consumption are obtained. Besides, the lightweight network has realized better accuracy than YOLOv5m. In conclusion, the method proposed in this paper is effective in real-time fish detection and recognition with practical application prospects.

This paper has addressed the terrain-following problem of an autonomous underwater vehicle for widely used ocean survey missions. Considering the terrain feature description with limited sensing ability in underwater scenarios, a vertically installed multi-beam sonar and a downward single-beam echo sounder are equipped to obtain seafloor detecting data online, and a local polynomial fitting algorithm is carried out with a receding horizon strategy in order to generate a proper tracking path to keep the desired height above the sea bottom. With the construction of the autonomous underwater vehicle dynamic model in the North East Down frame regarding the vertical plane, an online sparse identification algorithm is implemented to obtain the model parameters during the diving process. Then, a fractional-order sliding mode controller is proposed to enable accurate tracking of the path planned and Lyapunov-based theory is utilized to prove the stability of the control algorithm. With the simulation results, the tracking effectiveness of the fractional-order sliding mode controller with in situ identification is verified.

This paper investigates the terrain-following problem for an autonomous underwater vehicle (AUV) from control perspectives with full consideration of terrain characteristics. By equipping the AUV with three simple single-beam echo sounders, a set of precise along-track bottom slopes are obtained in a real-time manner, and the occurrence of the lost bottom lock phenomenon caused by a single sonar altimeter is eliminated. A slope-based data processing method is developed, which enables an AUV to characterize seafloor features with complementary sensing modalities to generate proper adaptive height-modified values. In order to keep a fixed height when maneuvering over rugged terrains, a back-stepping depth control is implemented by adjusting horizontal rudder angle, and Lyapunov theory is adopted to analyze the asymptotic stability of the resulting terrain-following control system. At last, simulation results indicate the feasibility and effectiveness of the proposed methodology with a discussion of various sensor configurations.