Sea-land segmentation of optical remote sensing images holds great importance for military and civilian applications, such as coastal monitoring, target detection, and resource management. Although convolutional neural networks (CNNs) have achieved significant improvements in semantic segmentation, the challenges of efficient feature extraction, representation, fusion, and information transmission remain unsolved, which especially impacts the segmentation effectiveness of existing CNN-based models in extracting irregular and refined sea-land boundaries. In this paper, a novel Dual-Branch Ensemble Network (DBENet) is proposed for pixel-level sea-land segmentation. The salient properties of the DBENet are: 1) A novel dual-branch network architecture is developed to achieve sufficient feature extraction and representation. 2) An efficient ensemble attention learning strategy suitable for the DBENet is designed to strengthen the correlation between dual branches to further facilitate feature fusion and information transmission. The comparative study with state-of-the-art methods reveals the superior performance of our approach, and the ablation study demonstrates the effectiveness of each component in the proposed network. The source code is available at https://github.com/RobertTang0/DBENet.
13 May 2024 in Engineering Applications of Artificial Intelligence
Underwater object detection (UOD) has attracted widespread attention, being of great significance for marine resource management, underwater security and defense, underwater infrastructure inspection, etc. However, high-quality UOD tasks often encounter challenges such as image quality degradation, complex backgrounds, and occlusions between objects at different scales. This paper presents a collaborative framework for UOD via joint image enhancement and super-resolution to address the above problems. Specifically, a joint-oriented framework is constructed incorporating underwater image enhancement and super-resolution techniques. The proposed framework is capable of generating a detection-favoring appearance to provide more visual cues for UOD tasks. Furthermore, a plug-and-play self-attention mechanism, termed multihead blurpooling fusion network (MBFNet), is developed to capture sufficient contextual information by focusing on the dependencies between multiscale feature maps, so that the UOD performance of our proposed framework can be further facilitated. A comparative study on the popular URPC2020 and Brackish datasets demonstrates the superior performance of our proposed collaborative framework, and the ablation study also validates the effectiveness of each component within the framework.
23 September 2025 in IEEE Transactions on Broadcasting
Underwater image enhancement (UIE) aims to provide high-quality observations of challenging underwater scenarios, which is of great significance for various broadcast technologies. Extensive non-learning-based and learning-based UIE methods have been presented and applied. However, non-learning-based strategies typically struggle to demonstrate superior generalization capabilities, while learning-based strategies generally suffer from potential over- or under-enhancement due to the lack of sufficient prior knowledge. To address the challenges above, this paper presents a heuristic cooperative-competitive network, termed Co2Net. Specifically, our Co2Net integrates non-learning mechanisms into the deep learning framework to achieve information fusion from explainable prior knowledge and discernible hierarchical features, thereby facilitating promising and reasonable enhancement of degraded underwater images. Furthermore, our Co2Net adopts a hybrid convolutional neural network (CNN)-Transformer architecture, which comprises successive cooperative-competitive modules (Co2Ms) to achieve adequate extraction, representation, and transmission of both prior knowledge and discernible features. Comprehensive experiments are conducted to demonstrate the superiority and universality of our proposed Co2Net, and sufficient ablation studies are also performed to reveal the effectiveness of each component within our model. The source code is available at https://github.com/jixun-dmu/Co2Net