Gaze behavior is important and non-invasive human–computer interaction information that plays an important role in many fields—including skills transfer, psychology, and human–computer interaction. Recently, improving the performance of appearance-based gaze estimation, using deep learning techniques, has attracted increasing attention: however, several key problems in these deep-learning-based gaze estimation methods remain. Firstly, the feature fusion stage is not fully considered: existing methods simply concatenate the different obtained features into one feature, without considering their internal relationship. Secondly, dynamic features can be difficult to learn, because of the unstable extraction process of ambiguously defined dynamic features. In this study, we propose a novel method to consider feature fusion and dynamic feature extraction problems. We propose the static transformer module (STM), which uses a multi-head self-attention mechanism to fuse fine-grained eye features and coarse-grained facial features. Additionally, we propose an innovative recurrent neural network (RNN) cell—that is, the temporal differential module (TDM)—which can be used to extract dynamic features. We integrated the STM and the TDM into the static transformer with a temporal differential network (STTDN). We evaluated the STTDN performance, using two publicly available datasets (MPIIFaceGaze and Eyediap), and demonstrated the effectiveness of the STM and the TDM. Our results show that the proposed STTDN outperformed state-of-the-art methods, including that of Eyediap (by 2.9%).

With the large-scale development of smart grids and the integration of renewable energy, the operational complexity and load volatility of power systems have increased significantly, placing higher demands on the accuracy and timeliness of electricity load forecasting. However, existing methods struggle to capture the nonlinear and volatile characteristics of load sequences, often exhibiting insufficient fitting and poor generalization in peak and abrupt change scenarios. To address these challenges, this paper proposes a deep learning model named CGA-LoadNet, which integrates a one-dimensional convolutional neural network (1D-CNN), gated recurrent units (GRUs), and a self-attention mechanism. The model is capable of simultaneously extracting local temporal features and long-term dependencies. To validate its effectiveness, we conducted experiments on a publicly available electricity load dataset. The experimental results demonstrate that CGA-LoadNet significantly outperforms baseline models, achieving the best performance on key metrics with an $R^2$ of 0.993, RMSE of 18.44, MAE of 13.94, and MAPE of 1.72, thereby confirming the effectiveness and practical potential of its architectural design. Overall, CGA-LoadNet more accurately fits actual load curves, particularly in complex regions, such as load peaks and abrupt changes, providing an efficient and robust solution for short-term load forecasting in smart grid scenarios.

Metasurfaces have the ability to manipulate electromagnetic waves, which allows for the creation of functions such as perfect absorbers. The goal of a perfect absorber is to achieve high absorption peaks within a specific frequency band. This paper introduces an improved metasurface absorber structure that can achieve efficient absorption in four different frequency bands within the range of 2-9 GHz. In the field of metasurface design, deep learning methods have been recently applied due to their powerful data processing capabilities. However, these methods have primarily used fully connected neural networks and Long Short-Term Memory (LSTM). Despite their capabilities, fully connected networks and LSTM struggle to capture the global information in absorption spectrum data, leading to less accurate predictions. In this study, it was observed that the Transformer model can effectively capture global information using Multi-Head Self-Attention (MHSA) and is not affected by the length of the data. Based on this observation, this paper presents a lightweight model consisting solely of an encoder, achieving a Mean Squared Error (MSE) that is one-twentieth of the State-of-the-Art (SOTA). This model predicts metasurface structure based on target absorption spectra, enabling users to rapidly obtain metasurface absorber structures directly from input absorption spectra. The model consists of two parts: embedding and encoder. The embedding processes input absorption spectra data and adds positional encoding, while the encoder extracts spectral data features. MHSA effectively captures contextual information of absorption spectra, emphasizing key feature information. The final model achieved a MSE convergence of 2 × 10⁻⁴ and a coefficient of determination (R ²)value of 0.998, successfully optimizing the design of multi-band metasurface absorbers. Moreover, the predicted results from the model exhibit an absorption spectrum that is highly consistent with the target spectrum.

With the growing use of drones, efficient detection algorithms are crucial, especially under adverse weather conditions. Most existing drone detection algorithms perform well only in clear weather, resulting in significant performance drops in foggy conditions. This study focuses on improving drone detection in foggy environments using the Mean Teacher framework for domain adaptation. The Mean Teacher framework’s performance relies on the quality of the teacher model’s pseudo-labels. To enhance the quality of the pseudo-labels from the teacher model, we introduce Foggy Drone Teacher (FDT), which includes three key components: (1) Adaptive Style and Context Augmentation to reduce domain shift and improve pseudo-label quality; (2) Simplified Domain Alignment with a novel adversarial strategy to boost domain adaptation; and (3) Progressive Domain Adaptation Training, a two-stage process that helps the teacher model produce more stable and accurate pseudo-labels. In addition, owing to the lack of publicly available data, we created Foggy Drone Dataset (FDD) to support this research. Extensive experiments show that our model achieves a 21.1-point increase in $A{P}_{0.5}$ compared to the baseline and outperforms state-of-the-art models. This method significantly improves drone detection accuracy in foggy conditions.