Facial expression recognition (FER) becomes more challenging in the wild due to unconstrained conditions, such as the different illumination, pose changes, and occlusion of the face. Current FER methods deploy the attention mechanism in deep neural networks to improve the performance. However, these models only capture the limited attention features and relationships. Thus this paper proposes a novel FER framework called multi-relations aware network (MRAN), which can focus on global and local attention features and learn the multi-level relationships among local regions, between global-local features and among different samples, to obtain efficient emotional features. Specifically, our method first imposes the spatial attention on both the whole face and local regions to simultaneously learn the global and local salient features. After that, a region relation transformer is deployed to capture the internal structure among local facial regions, and a global-local relation transformer is designed to learn the fusion relations between global features and local features for different facial expressions. Subsequently, a sample relation transformer is deployed to focus on intrinsic similarity relationship among training samples, which promotes invariant feature learning for each expression. Finally, a joint optimization strategy is designed to efficiently optimize the model. The conducted experimental results on in-the-wild databases show that our method obtains the superior performance compared to some state-of-the-art models.

Facial expression recognition has been widely used to solve the problems such as lie detection and human-machine interaction. However, due to the difficulties to control the application environments, current methods have the lower recognition accuracy in practice. This paper proposes a new method for facial expression recognition by considering several aspects. First, human beings are easy to recognize some expressions, while difficult to recognize others. Inspired by this intuition, a new loss function is proposed to enlarge the distances between samples from easily confused categories. Second, human learning is divided into many stages, and the learning objective of each stage is different. Thus, dynamic objectives learning is proposed, where each objective at different stage is defined by the corresponding loss function. In order to better realize the above ideas, a new deep neural network for facial expression recognition is proposed, which integrates the covariance pooling layer and residual network units into the deep convolution neural network so as to better perform dynamic objectives learning. The experimental results on the standard databases verify the effectiveness and the superior performance of our methods.

To suppress the spread of COVID-19, accurate diagnosis at an early stage is crucial, chest screening with radiography imaging plays an important role in addition to the real-time reverse transcriptase polymerase chain reaction (RT-PCR) swab test. Due to the limited data, existing models suffer from incapable feature extraction and poor network convergence and optimization. Accordingly, a multi-stage residual network, MSRCovXNet, is proposed for effective detection of COVID-19 from chest x-ray (CXR) images. As a shallow yet effective classifier with the ResNet-18 as the feature extractor, MSRCovXNet is optimized by fusing two proposed feature enhancement modules (FEM), i.e., low-level and high-level feature maps (LLFMs and HLFMs), which contain respectively more local information and rich semantic information, respectively. For effective fusion of these two features, a single-stage FEM (MSFEM) and a multi-stage FEM (MSFEM) are proposed to enhance the semantic feature representation of the LLFMs and the local feature representation of the HLFMs, respectively. Without ensembling other deep learning models, our MSRCovXNet has a precision of 98.9% and a recall of 94% in detection of COVID-19, which outperforms several state-of-the-art models. When evaluated on the COVIDGR dataset, an average accuracy of 82.2% is achieved, leading other methods by at least 1.2%.

Semantic segmentation of remote sensing images has significant advances with the adoption of deep neural networks, taking the advantages of Convolutional Neural Networks (CNNs) in local feature extraction with Transformers in global information modeling. However, due to the limitations of CNNs in long-range modeling capabilities and the computational complexity constraints of Transformers, remote sensing semantic segmentation still faces issues such as serious holes, rough edge segmentation, false and even missed detections caused by the light, shadow and other factors. To address these issues, we propose a visual state space model called GLVMamba, which employs CNNs as the encoder and the proposed Global-Local Visual State Space (GLVSS) block as the core decoder. Specifically, the GLVSS block introduces locality forward feedback and shift window mechanism to addresses the deficiency of insufficient modeling of neighboring pixel dependencies of Mamba, which enhances the integration of global and local context during feature reconstruction, boosts object perception capabilities of the model, and effectively refines edge contours. Additionally, the scale-aware pyramid pooling (SCPP) module is proposed to fully merge the features from various scales and adaptively fuse and extract the distinguishing features to mitigate the holes and false detections. The GLVMamba effectively captures global-local semantic information and multi-scale feature through the GLVSS block and the SCPP module, achieving efficient and accurate remote sensing semantic segmentation. Extensive experiments on two widely used datasets have effectively demonstrated the superiority of our proposed method over the other state-of-the-art methods. The code will be available at https://github.com/Tokisakiwlp/GLVMamba.

Underwater object detection (UOD) presents substantial challenges due to the complex visual conditions and the physical properties of light in underwater environments. Small aquatic creatures often congregate in large groups, further complicating the task. To address these challenges, we develop Aqua-DETR, a tailored end-to-end framework for UOD. Our method includes an align-split network to enhance multi-scale feature interaction and fusion for small object identification and a distinction enhancement module using various attention mechanisms to improve ambiguous object identification. Experimental results on four challenging datasets demonstrate that Aqua-DETR outperforms most existing state-of-the-art methods in the UOD task, validating its effectiveness and robustness.