Accurate image reconstruction from low-sampling measurements remains a crucial challenge in image compressed sensing (CS). Image CS has attracted increasing research interest, with deep learning approaches being extensively studied. Deep learning based methods reduce computational complexity at the inference stage compared with traditional iterative solvers and can improve reconstruction quality. In particular, methods built on transformer architectures have achieved remarkable success in tasks involving image reconstruction and restoration. However, as the depth of the model increases, some content information may be attenuated during feature propagation. To alleviate the issue, this paper proposes a content-aware transformer network (CATNet) for image compressed sensing. It focuses on content feed-forward and channel self-attention to effectively obtain sufficient feature representation. In the initial reconstruction stage, this paper employs trainable sensing matrix to adaptively capture sparse structures from the image, which helps improve measurement efficiency. In the deep reconstruction stage, this paper introduces the content-focused feed-forward network (CFFN) to strengthen the effective content features. Besides, this paper employs the channel-focused self-attention (CFSA) to enhance global features and integrate convolution to refine local details. Finally, this paper adopts a multi-stage supervision to ensure sufficient training. Comprehensive experiments conducted on 5 widely used datasets demonstrate the strong performance and superiority of CATNet over 9 representative methods, especially under extremely low sampling rates. At CS ratio of 0.05, CATNet achieves competitive results, with PSNR/SSIM values of 22.25/0.6907 on Set5, 21.91/0.6623 on Set11, 21.99/0.6664 on Set14, 23.97/0.6964 on BSD68, and 20.61/0.5585 on Urban100. Moreover, the method has been successfully implemented in the self-developed single-pixel imaging optical system, further demonstrating its applicability.

Salient object detection in remote sensing images (RSI-SOD) plays a vital role in enabling large-scale environmental perception for embodied intelligence systems. Despite growing interest, existing methods often underperform in complex urban settings characterized by cluttered backgrounds and significant scale variations, suffering from either blurred object boundaries or high computational costs. To address these issues, we propose a perceptual gating mamba network (PGMNet) that accurately and efficiently detects objects with a global receptive field. The network first employs a spatial neighboring scanning strategy to extract structurally coherent image patches. A gated focus-aware mamba fusion module is then designed to enhance spatial coherence across these patches through joint channel focusing, spatial attention, and a gating mechanism, thereby improving localization accuracy. Furthermore, a structure-preserving upsampling aggregation module is developed to reconstruct complete object contours by aligning and aggregating multi-level features during decoding. Extensive experiments on two benchmark datasets demonstrate that PGMNet outperforms 12 competing methods, validating its effectiveness and robustness in complex urban environments.

Vehicle object detection is a fundamental component of intelligent transportation systems, yet its deployment in resource-constrained environments is limited by the trade-off between accuracy and computational efficiency. This work systematically examines the relationship between network depth and detection performance through controlled experiments on both convolutional and attention-based architectures. The results reveal that accuracy does not scale indefinitely with depth, but instead exhibits diminishing returns and converges to a task-dependent performance plateau. A mutual information-based analysis attributes this saturation to fundamental limitations in information retention and representational efficiency. Motivated by this insight, we propose the TritonFocus Network (TF-Net), an architecture that integrates large-Kernel contextual perception with small-Kernel local aggregation for efficient feature representation. Extensive evaluations on the UA-DETRAC and BITVehicle benchmarks demonstrate that TF-Net achieves a state-of-the-art accuracy–efficiency trade-off among lightweight detectors. TF-Net achieves a relative detection accuracy improvement of approximately 5.8%, while maintaining high computational efficiency and approaching twice the inference speed of the baseline model. These findings validate the efficacy of saturation-aware architectural optimization and provide theoretical and practical guidance for deploying high-performance detection models in edge computing scenarios.

Salient object detection in optical remote sensing images (RSI-SOD) is a newly developing task and has attracted extensive attention. Although some satisfactory performances have been achieved, there are still some poor performance situations, such as complex backgrounds and objects with various scales. Because the existing methods have the issue of insufficient adjacent feature representation and inappropriate constraint supervision. To alleviate the issues, we attempt to utilize the adjacent complementary and introduce the global constraints. Accordingly, we propose the adjacent complementary network (ACNet) with global constraints for RSI-SOD. Specifically, in the encoder, we mainly design the adjacent complementary module (ACM) to complement the features among the current layer and adjacent layers. Therefore, it can effectively aggregate saliency features. Besides, we have equipped a global average pooling (GAP) to enhance global information. In the decoder, we design a fusion enhancement module (FEM) to enhance the feature conversion between the coding layer and decoding layer. Finally, we employ a novel loss, which introduces global constraints and can supervise the objects with various scales. Extensive studies on two benchmark datasets demonstrate that ACNet is competitive and outperforms the 15 state-of-the-art (SOTA) approaches.

At present, most saliency detection methods are based on fully convolutional neural networks (FCNs). However, FCNs usually blur the edges of salient objects. Due to that, the multiple convolution and pooling operations of the FCNs will limit the spatial resolution of the feature maps. To alleviate this issue and obtain accurate edges, we propose a hierarchical edge refinement network (HERNet) for accurate saliency detection. In detail, the HERNet is mainly composed of a saliency prediction network and an edge preserving network. Firstly, the saliency prediction network is used to roughly detect the regions of salient objects and is based on a modified U-Net structure. Then, the edge preserving network is used to accurately detect the edges of salient objects, and this network is mainly composed of the atrous spatial pyramid pooling (ASPP) module. Different from the previous indiscriminate supervision strategy, we adopt a new one-to-one hierarchical supervision strategy to supervise the different outputs of the entire network. Experimental results on five traditional benchmark datasets demonstrate that the proposed HERNet performs well when compared with the state-of-the-art methods.

Convolutional Neural Network (CNN) has shown their advantages in salient object detection. CNN can generate great saliency maps because it can obtain high-level semantic information. And the semantic information is usually achieved by stacking multiple convolutional layers and pooling layers. However, multiple pooling operations will reduce the size of the feature map and easily blur the boundary of the salient object. Therefore, such operations are not beneficial to generate great saliency results. To alleviate this issue, we propose a novel edge information-guided hierarchical feature fusion network (HFFNet). Our network fuses features hierarchically and retains accurate semantic information and clear edge information effectively. Specifically, we extract image features from different levels of VGG. Then, we fuse the features hierarchically to generate high-level semantic information and low-level edge information. In order to retain better information at different levels, we adopt a one-to-one hierarchical supervision strategy to supervise the generation of low-level information and high-level information respectively. Finally, we use low-level edge information to guide the saliency map generation, and the edge guidance fusion is able to identify saliency regions effectively. The proposed HFFNet has been extensively evaluated on five traditional benchmark datasets. The experimental results demonstrate that the proposed model is fairly effective in salient object detection compared with 10 state-of-the-art models under different evaluation indicators, and it is superior to most of the comparison models.