Marine debris impacts negatively upon the marine environment and the survival of marine life because they are some difficult-to-degrade substances, and most of them will sink into the deep sea and continue to exist in the ocean. Autonomous underwater vehicles can clean up the deep-sea debris to some extent. However, the efficient detection method plays a critical role in the collection rate. This article establishes an efficient deep-sea debris detection method with high speed using deep learning methods. First, a real deep-sea debris detection dataset (3-D dataset) is established for further research. The dataset contains seven types of debris: cloth, fishing net and rope, glass, metal, natural debris, rubber, and plastic. Second, the one-stage deep-sea debris detection network ResNet50-YOLOV3 is proposed. In addition, eight advanced detection models are also involved in the detection process of deep-sea debris. Finally, the performance of ResNet50-YOLOV3 is verified by experiments. Furthermore, the applicability and effectiveness of ResNet50-YOLOV3 in deep-sea debris detection are proved by the experimental results.

In contemporary society full of stereoscopic images, how to assess visual quality of 3D images has attracted an increasing attention in field of Stereoscopic Image Quality Assessment (SIQA). Compared with 2D-IQA, SIQA is more challenging because some complicated features of Human Visual System (HVS), such as binocular interaction and binocular fusion, must be considered. In this paper, considering both binocular interaction and fusion mechanisms of the HVS, a hierarchical no-reference stereoscopic image quality assessment network (StereoIF-Net) is proposed to simulate the whole quality perception of 3D visual signals in human cortex, including two key modules: BIM and BFM. In particular, Binocular Interaction Modules (BIMs) are constructed to simulate binocular interaction in V2-V5 visual cortex regions, in which a novel cross convolution is designed to explore the interaction details in each region. In the BIMs, different output channel numbers are designed to imitate various receptive fields in V2-V5. Furthermore, a Binocular Fusion Module (BFM) with automatic learned weights is proposed to model binocular fusion of the HVS in higher cortex layers. The verification experiments are conducted on the LIVE 3D, IVC and Waterloo-IVC SIQA databases and three indices including PLCC, SROCC and RMSE are employed to evaluate the assessment consistency between StereoIF-Net and the HVS. The proposed StereoIF-Net achieves almost the best results compared with advanced SIQA methods. Specifically, the metric values on LIVE 3D, IVC and WIVC-I are the best, and are the second-best on the WIVC-II.

Deep-sea debris is a globally growing problem, which is negatively impacting biological and chemical ecosystems. More seriously, the debris is likely to persist in the deep sea for long periods. Fortunately, with the help of the debris detection system the submersibles can clean up the debris. An excellent classifier is critical to the debris detection system. Therefore, the objective of this study is to determine whether deep convolutional neural networks can distinguish the differences of debris and natural deep-sea environment, so as to effectively achieve deep-sea debris identification. First, a real deep-sea debris images dataset is constructed for further classification research based on an online deep-sea debris database owned by the Japan Agency for Marine-Earth Science and Technology. Second, the hybrid Shuffle-Xception network is constructed to classify the deep-sea image as metal, glass, plastic, rubber, fishing net & rope, natural debris, and cloth. Furthermore, five common convolutional neural networks (CNNs) frameworks are also employed to implement the classification process. Finally, the identification experiments are carried out to validate the performance of the proposed methodology. The results demonstrate that the proposed method is superior to the state-of-the-art CNN method and has the potential for deep-sea debris identification.

Oceanic eddies have a non-negligible impact on ocean energy transfer, nutrient distribution, and biological migration in global oceans. The fine detection of oceanic eddies is significant for the development of marine science. Remarkable achievements of eddy recognition were achieved by mining the satellite altimeter data and its derived data. However, due to the limited spatial resolution of the altimeters, it is difficult to detect the submesoscale oceanic eddies with radial dimensions less than 10 km. Different from the previous works, the context and edge association network (CEA-Net) is proposed to identify submesoscale oceanic eddies with high spatial resolution Sentinel-1 data. The edge information fusion module (EIFM) is designed to associate the context and edge feature more accurately and efficiently. Furthermore, a multi-scale eddy detection strategy is proposed and applied to Sentinel-1 interferometric wide swath data to solve the scale problem of oceanic eddy detection. Specifically, a manually interpreted dataset, SAR-Eddy 2019, was constructed to address the dilemma of insufficient datasets for submesoscale oceanic eddy detection. The experimental results demonstrate that CEA-Net can outperform other mainstream models with the highest mAP reaching 85.47% with SAR-Eddy 2019 dataset. The CEA-Net proposed in this research provides important significance for the study of submesoscale oceanic eddies.