Salient Object Detection (SOD) in RGB-D images plays a crucial role in the field of computer vision, with its central aim being to identify and segment the most visually striking objects within a scene. However, optimizing the fusion of multi-modal and multi-scale features to enhance detection performance remains a challenge. To address this issue, we propose a network model based on semantic localization and multi-scale fusion (SLMSF-Net), specifically designed for RGB-D SOD. Firstly, we designed a Deep Attention Module (DAM), which extracts valuable depth feature information from both channel and spatial perspectives and efficiently merges it with RGB features. Subsequently, a Semantic Localization Module (SLM) is introduced to enhance the top-level modality fusion features, enabling the precise localization of salient objects. Finally, a Multi-Scale Fusion Module (MSF) is employed to perform inverse decoding on the modality fusion features, thus restoring the detailed information of the objects and generating high-precision saliency maps. Our approach has been validated across six RGB-D salient object detection datasets. The experimental results indicate an improvement of 0.20~1.80%, 0.09~1.46%, 0.19~1.05%, and 0.0002~0.0062, respectively in maxF, maxE, S, and MAE metrics, compared to the best competing methods (AFNet, DCMF, and C2DFNet).
In the field of Visual Saliency Detection, accurately segmenting salient objects from images is crucial for various applications such as image editing and visual tracking. However, this task becomes particularly challenging due to several issues encountered across different scenes, including (i) the high diversity of salient objects in terms of size, color, and texture; and (ii) the unclear boundaries between salient objects and their backgrounds. To address these challenges, this paper introduces a deep convolutional network based on global guidance map and background attention, named GMBA-Net, for precise segmentation of salient objects. Initially, the Cross-Modal Multi-Scale Attention Module (CMAM) finely integrates information from RGB and depth images. This module employs attention mechanisms to enhance the complementarity between modalities and the expressiveness of multi-scale features, thereby capturing saliency information more effectively at different scales. Following this, the Interactive Global Context Module (IGCM) is utilized to generate a global guidance map, which leverages semantic information from high-level features for accurate localization of salient regions. Finally, a background attention mechanism (BGAM) refines the boundaries of salient objects, significantly improving the clarity of the edges in the saliency maps and achieving precise delineation of salient object boundaries. Experiments conducted on multiple public RGB-D saliency detection datasets have validated the effectiveness of our proposed method. Compared to existing state-of-the-art methods, our model demonstrates significant improvements in terms of saliency detection accuracy and boundary clarity, substantially enhancing the overall performance of saliency detection.
The Information Fidelity Criterion (IFC) of image assessment metric is based on the statistical mutual information shared by reference image and distorted image. IFC is parameterless and superior to traditional methods. However, human vision system (HVS) is adaptive to different scale feature of image. A multi-scale IFC approach to model the characteristics of HVS was proposed in this paper, in which image was locally filtered by Gaussian dynamic windows ratio to different scales. Finally, assessment scores for different scales were pooled with weight of contrast sensitivity function. Experimental results showed that the performance of the proposed approach was superior to other methods.
Real-time path planning for Unmanned Surface Vehicles (USVs) in complex marine environments remains challenging due to unstructured environments, ocean current disturbances, and dynamic obstacles. This paper proposes an improved Hybrid Safety and Reward-Sensitive Twin Delayed Deep Deterministic Policy Gradient (H_RS_TD3) algorithm and constructs a high-fidelity simulation environment based on GEBCO bathymetric data and CMEMS ocean current data. The path planning problem is formulated as a Markov Decision Process (MDP), where the state space incorporates multi-beam radar perception, ocean current disturbances, and relative goal information, while the action space outputs continuous thrust and rudder commands subject to vehicle dynamics constraints. The proposed framework integrates a risk-aware hybrid safety decision architecture, a Trajectory Predictor Network (TPN), a Curvature-driven Advantage-based Prioritized Experience Replay (CDA-PER) mechanism, and an uncertainty-aware conservative Q-learning strategy to enhance navigation safety, sample efficiency, and policy stability. Comprehensive simulations demonstrate that, compared with baseline deep reinforcement learning methods, the proposed approach achieves faster convergence, improved stability, and competitive path efficiency while consistently maintaining sufficient obstacle clearance and millisecond-level inference latency, validating its effectiveness and practical feasibility for safe USV navigation in realistic dynamic marine environments.
Deep unfolding networks (DUNs) have attracted growing attention in compressed sensing (CS) due to their good interpretability and high performance. However, many DUNs often improve the reconstruction effect at the price of a large number of parameters and have the problem of feature information loss during iteration. This paper proposes a novel adaptive memory-augmented unfolding network for compressed sensing (AMAUN-CS). Concretely, without loss of interpretability, we integrate an adaptive content-aware strategy into the gradient descent step of the proximal gradient descent (PGD) algorithm, driving it to adaptively capture the adequate features. In addition, we extended AMAUN-CS based on the memory storage mechanism of the human brain and propose AMAUN-CS+ to develop the dependency of deep information across cascading stages. The experimental results show that the AMAUN-CS model surpasses other advanced methods on various public benchmark datasets while having lower complexity in training.
Polyp segmentation is vital for early detection and treatment of colorectal cancer, significantly improving patient prognosis. This paper proposes an efficient and precise polyp segmentation model called the Partial Decoder Localization and Foreground-Background Refinement Network (PDLFBR-Net), which simulates the human object recognition process. Specifically, PDLFBR-Net comprises three key modules: the Cross-level Attention-enhanced Fusion Module (CAFM), the Position Recognition Module (PRM), and the Foreground-Background Refinement Module (FBRM). The CAFM enhances feature representation by fusing information from adjacent levels, providing more discriminative features. The PRM module simulates the human recognition process by using a partial decoder to locate potential polyp tissues from a global perspective. Subsequently, the FBRM is used to perform specific recognition, gradually refining the initial prediction results through foreground and background focusing to achieve precise recognition. Extensive experiments demonstrate that the proposed PDLFBR-Net model significantly outperforms existing state-of-the-art models on five challenging datasets. On the Kvasir-SEG benchmark dataset, the mean Dice and mean IoU values reached 93.7% and 89.5%, respectively, which represents an improvement of 0.4% and 0.6% compared to the best-performing state-of-the-art (SOTA) method.