In practical application, facial image recognition is vulnerable to be attacked by photos, videos, etc., while some currently used artificial feature extractors in machine learning, such as activity detection, texture descriptors, and distortion detection, are insufficient due to their weak detection ability in feature extraction from unknown attack. In order to deal with the aforementioned deficiency and improve the network security, this paper proposes directional difference convolution for the deep learning in gradient image information extraction, which analyzes pixel correlation within the convolution domain and calculates pixel gradients through difference calculation. Its combination with traditional convolution can be optimized by a parameter $\theta$. Its stronger ability in gradient extraction improves the learning and predicting ability of the network, whose performance testing on CASIA-MFSD, Replay-Attack, and MSU-MFSD for face anti-spoofing task shows that our method outperforms the current related methods.

In dynamic and unstructured environments, the obstacle avoidance capabilities of Unmanned Aerial Vehicles (UAVs) are crucial for mission success. Traditional methods struggle with adaptability and effectiveness in unknown or changing scenes. In contrast, the commonly used deep reinforcement learning (DRL) ones suffer from slow convergence, reduced accuracy, and inadequate robustness due to the loss of sensitivity to outliers and parameter rigidity. To address these challenges, we propose an enhanced DRL framework that leverages a Dynamic Huber loss function tailored for UAV autonomous obstacle avoidance. By incorporating Soft updates for target network and dynamically tuning the Huber loss, the proposed method facilitates faster model convergence, superior control precision, and improved robustness. Both theoretical analysis and experimental simulation verify its effectiveness with superior planning success rate, shorter average path length, and faster model convergence over traditional approaches. Specifically, in static environments, the Dynamic Huber-loss-based DRL framework achieves a 98.85% success rate with an optimized average path length of 10.73; in dynamic environments, it attains a 74.20% success rate with an average path length of 37.04; adding wind disturbances in a dynamic environment, it attains a 70.95% success rate with an average path length of 40.40, highlighting its enhanced performance and adaptability.

Accurate detection of trading actions (buy, sell, and hold) is critical for portfolio optimization and risk management in volatile stock markets. However, existing approaches often suffer from deficiencies in feature representation, spatiotemporal modeling, and class balancing, which limit their effectiveness. To address these issues, we propose HA-CL, a deep learning framework that integrates a hierarchical attention mechanism with CNN-LSTM. Specifically, technical indicators are encoded into a structured 2D matrix to preserve the inherent characteristics of stocks. Features extracted by ResNet are processed by a channel-wise LSTM equipped with an attention core to adaptively fuse spatial, temporal, and channel-level importance. To mitigate class imbalance, we design a customized extrema labeling strategy augmented with extrema oversampling, an importance-aware focal loss, and a heuristic action recalibration. Experiments on 63 Chinese A-share stocks show that HA-CL achieves an average accuracy of 68.89% with an annualized return of 111.01%, substantially outperforming all baselines. Risk-adjusted return metrics such as the Sharpe Ratio and the Maximum Drawdown further validate its robustness across market conditions. Together, they highlight the potential of HA-CL to translate complex market patterns into profitable trading actions.

With the development of data mining technology, the analysis of event-related potential (ERP) data has evolved from statistical analysis of time-domain features to data-driven techniques based on supervised and unsupervised learning. However, there are still many challenges in understanding the relationship between ERP components and the representation of familiar and unfamiliar faces. To address this, this paper proposes a model based on Dynamic Multi-Scale Convolution for group recognition of familiar and unfamiliar faces. This approach uses generated weight masks for cross-subject familiar/unfamiliar face recognition using a multi-scale model. The model employs a variable-length filter generator to dynamically determine the optimal filter length for time-series samples, thereby capturing features at different time scales. Comparative experiments are conducted to evaluate the model’s performance against SOTA models. The results demonstrate that our model achieves impressive outcomes, with a balanced accuracy rate of 93.20% and an F1 score of 88.54%, outperforming the methods used for comparison. The ERP data extracted from different time regions in the model can also provide data-driven technical support for research based on the representation of different ERP components.