The underwater environment is still unknown for humans, so the high definition camera is an important tool for data acquisition at short distances underwater. Due to insufficient power, the image data collected by underwater submersible devices cannot be analyzed in real time. Based on the characteristics of Field-Programmable Gate Array (FPGA), low power consumption, strong computing capability, and high flexibility, we design an embedded FPGA image recognition system on Convolutional Neural Network (CNN). By using two technologies of FPGA, parallelism and pipeline, the parallelization of multi-depth convolution operations is realized. In the experimental phase, we collect and segment the images from underwater video recorded by the submersible. Next, we join the tags with the images to build the training set. The test results show that the proposed FPGA system achieves the same accuracy as the workstation, and we get a frame rate at 25 FPS with the resolution of 1920 × 1080. This meets our needs for underwater identification tasks.

Target search for moving and invisible objects has always been considered a challenge, as the floating objects drift with the flows. This study focuses on target search by multiple autonomous underwater vehicles (AUV) and investigates a multi-agent target search method (MATSMI) for moving and invisible objects. In the MATSMI algorithm, based on the multi-agent deep deterministic policy gradient (MADDPG) method, we add spatial and temporal information to the reinforcement learning state and set up specialized rewards in conjunction with a maritime target search scenario. Additionally, we construct a simulation environment to simulate a multi-AUV search for the floating object. The simulation results show that the MATSMI method has about 20% higher search success rate and about 70 steps shorter search time than the traditional search method. In addition, the MATSMI method converges faster than the MADDPG method. This paper provides a novel and effective method for solving the maritime target search problem.

Skeleton-based action recognition methods are limited by the semantic extraction of spatio-temporal skeletal maps. However, current methods have difficulty in effectively combining features from both temporal and spatial graph dimensions and tend to be thick on one side and thin on the other. In this paper, we propose a Temporal-Channel Aggregation Graph Convolutional Networks (TCA-GCN) to learn spatial and temporal topologies dynamically and efficiently aggregate topological features in different temporal and channel dimensions for skeleton-based action recognition. We use the Temporal Aggregation module to learn temporal dimensional features and the Channel Aggregation module to efficiently combine spatial dynamic topological features learned using Channel-wise with temporal dynamic topological features. In addition, we extract multi-scale skeletal features on temporal modeling and fuse them with priori skeletal knowledge with an attention mechanism. Extensive experiments show that our model results outperform state-of-the-art methods on the NTU RGB+D, NTU RGB+D 120, and NW-UCLA datasets.

The marine environment presents a unique set of challenges for human–robot interaction. Communicating with gestures is a common way for interacting between the diver and autonomous underwater vehicles (AUVs). However, underwater gesture recognition is a challenging visual task for AUVs due to light refraction and wavelength color attenuation issues. Current gesture recognition methods classify the whole image directly or locate the hand position first and then classify the hand features. Among these purely visual approaches, textual information is largely ignored. This paper proposes a visual–textual model for underwater hand gesture recognition (VT-UHGR). The VT-UHGR model encodes the underwater diver’s image as visual features, the category text as textual features, and generates visual–textual features through multimodal interactions. We guide AUVs to use image–text matching for learning and inference. The proposed method achieves better performance than most existing purely visual methods on the dataset CADDY, demonstrating the effectiveness of using textual patterns for underwater gesture recognition.

Subsea pipeline is the safest, most reliable, and most economical way to transport oil and gas from an offshore platform to an onshore terminal. However, the pipelines may rupture under the harsh working environment, causing oil and gas leakage. This calls for a proper device and method to detect the state of subsea pipelines in a timely and precise manner. The autonomous underwater vehicle carrying side-scan sonar offers a desirable way for target detection in the complex environment under the sea. As a result, this article combines the field-programmable gate array, featuring high throughput, low energy consumption and a high degree of parallelism, and the convolutional neural network into a sonar image recognition system. First, a training set was constructed by screening and splitting the sonar images collected by sensors, and labeled one by one. Next, the convolutional neural network model was trained by the set on the workstation platform. The trained model was integrated into the field-programmable gate array system and applied to recognize actual datasets. The recognition results were compared with those of the workstation platform. The comparison shows that the computational precision of the designed field-programmable gate array system based on convolutional neural network is equivalent to that of the workstation platform; however, the recognition time of the designed system can be saved by more than 77%, and its energy consumption can also be saved by more than 96.67%. Therefore, our system basically satisfies our demand for energy-efficient, real-time, and accurate recognition of sonar images.