In the digital media era, memes significantly influence emotional expression but can also convey hate through symbols or images, inciting harmful speech and division. The existing hateful memes detection methods are limited in their learning and generalization capabilities due to insufficient training data, which is caused by the difficulty of annotation. Therefore, we analyze the definition of memes from the perspective of social science and find that some representative visual features are repeatedly used as the carrier of hate metaphors in memes. Based on this, we propose a Representative Sample Augmented Hateful Memes Detection method (RSAHMD), which significantly improves the detection performance and generalizability of the model by introducing representative samples containing key visual features. Specifically, firstly, RSAHMD uses a representative sample retrieval method based on visual representative interpretation to select the most representative samples from existing datasets as auxiliary inputs, helping the model focus on metaphorical hate information. Secondly, we design an Adaptive Feature Weighting Module (AFWM) to dynamically adjust feature weights, enhancing the model's focus on key information while effectively reducing noise introduced from representative samples. Finally, we use the feedback mechanism to optimize the decision boundary of the model, and contrastive learning is applied to optimize the feature space distribution, reducing the interference of mislabeled samples and improving the ability of model to distinguish and retrieve complex samples. Experiments show RSAHMD improves performance and interpretability in detecting hateful memes. Its plug-and-play nature allows flexible application to different models and datasets, offering an efficient and transparent solution.

Deep learning models have made extraordinary progress in recent years, but image recognition in low-light conditions has not been studied much. Due to its application prospects, low-light image recognition has still received focuses. Low-light image recognition is still challenging because of the crucial feature information that is hard to mine and learned by a deep learning model. Compared with enhancing the image's brightness before recognition, recognizing low-light images by end-to-end mode straightly has a more practical sense. In this paper, we propose a novel end-to-end model named Attention-Capsule Network (ACNet) for low-light image recognition tasks. The proposed model is extend from capsule structure, and its key component is the Global-Local Attention (GLA) module. The GLA module is designed to effectively and comprehensively combine important information. It mines global perception information by global block and gets local detail information from the particular local union. This strategy can significantly improve the performance of the proposed model. In addition, this paper proposes a directional learning loss, which guides the model to extract key features of images by optimizing the error between reconstructed images and normal-light images. Our experimental results demonstrate the effectiveness and robustness of our proposed attention-capsule Network model for low-light image recognition.

Abstract Semantic segmentation is applied extensively in autonomous driving and intelligent transportation with methods that highly demand spatial and semantic information. Here, an STDC‐MA network is proposed to meet these demands. First, the STDC‐Seg structure is employed in STDC‐MA to ensure a lightweight and efficient structure. Subsequently, the feature alignment module is applied to understand the offset between high‐level and low‐level features, solving the problem of pixel offset related to upsampling on the high‐level feature map. The approach implements the effective fusion between high‐level features and low‐level features. A hierarchical multiscale attention mechanism is adopted to reveal the relationship among attention regions from two different input sizes of one image. Through this relationship, regions receiving much attention are integrated into the segmentation results, thereby reducing the unfocused regions of the input image and improving the effective utilisation of multiscale information. STDC‐MA maintains the segmentation speed as the STDC‐Seg network while improving the segmentation accuracy of small objects. STDC‐MA was verified on the validation dataset of Cityscapes. The segmentation result of STDC‐MA attained 78.32% mIOU with the input of 0.5× scale, 4.92% higher than STDC‐Seg.

Recent advances in single image super-resolution (SISR) demonstrate the power of deep learning for achieving better performance. Because it is costly to recollect the training data and retrain the model for infrared (IR) image super-resolution, the availability of only a few samples for restoring IR images presents an important challenge in the field of SISR. To solve this problem, we first propose the progressive super-resolution generative adversarial network (PSRGAN) that includes the main path and branch path. The depthwise residual block (DWRB) is used to represent the features of the IR image in the main path. Then, the novel shallow lightweight distillation residual block (SLDRB) is used to extract the features of the readily available visible image in the other path. Furthermore, inspired by transfer learning, we propose the multistage transfer learning strategy for bridging the gap between different high-dimensional feature spaces that can improve the PSGAN performance. Finally, quantitative and qualitative evaluations of two public datasets show that PSRGAN can achieve better results compared to the SR methods.

Recently, improved performance has been achieved in image super-resolution (SR) by using deep convolutional neural networks (CNNs). However, most existing networks neglect the feature correlations of adjacent layers, causing features at different levels to not be fully utilized. In this paper, a novel difference value network (DVN) is proposed to address this problem. The proposed network makes full use of different levels of features by using the difference values (D-values) of adjacent layers. Specifically, a difference value block (DVB) is designed to extract the difference values of adjacent layers. The extracted difference value can highlight which regions should be paid more attention to, so as to guide image SR. Further, a difference value group (DVG) is designed to integrate the difference values extracted by the difference value block into its output. In this way, the DVG can provide additional structure prior for image SR. Finally, to make our network more stable, a multipath supervised reconstruction block is proposed to supervise the reconstruction process. The experimental results on five benchmark datasets show that the proposed network can achieve better reconstruction results than the compared SR methods.

This paper proposes a data-driven-based optimization method for var-voltage sequential control (V2SC). First, power system var-voltage control characteristics, defined as the function between the reactive power injections and the bus voltages, are approximated by the var-voltage sensitivities (V2S). Then, V2S is estimated online using the noise-assisted ensemble regression method. Subsequently, an optimal model is proposed for V2SC based on the V2S estimation. To avoid frequent back and forth control actions, a hysteresis control strategy is employed. The performance of the proposed method is statistically validated in the 8-generator 36-node system and the Nordic32 system with test data measured from real power systems.