Abstract

(1) Background: Accurately decoding motor imagery (MI) tasks is a prerequisite for creating a MI-based brain-computer interface (BCI). However, low signal-to-noise ratio and non-stationarity of EEG signals present a huge challenge for the classification of MI-EEG signals, restricting the extensive development of the BCI industry.''' '''(2) Methods: In this paper, we propose a novel deep learning model CTANet that integrates both channel and temporal attention mechanisms into a convolutional neural network to improve the classification accuracy of the MI-BCI systems. The model is constituted first by three serially connected temporal, spatial, and temporal convolution layers to extract features from the brain signals, with an efficient channel attention module inserted between the second and the third convolutional layers to highlight useful feature channels. Subsequently, the time segment for task decoding is partitioned into several time windows, and a variance layer is employed for computing the logarithmic variance of each window. Next, a multi-head attention mechanism is adopted to extract temporal dependency of features from different windows. Finally, a fully connected layer is used for classifying MI-EEG signals. (3) Results: The performance of the proposed model was evaluated on two publicly available BCI datasets and compared with the state-of-the-art methods. The experimental results show that for dataset BCIC-IV2a, our network achieved classification accuracies of 81.17% and 84.33% in inter-session and intra-session scenarios respectively, whereas for dataset OpenBMI, our network achieved classification accuracy of 73.06% and 77.59% in inter-session and intra-session scenarios respectively. (4) Conclusions: These results outperform state-of-the-art networks, indicating significant potential of the proposed model CTANet in MI decoding.

Abstract

Accurate prediction of future vehicle trajectories is essential for ensuring safety and reliable decision-making in autonomous driving systems. However, existing deep learning-based approaches exhibit several limitations. Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs) struggle to effectively model long-term temporal dependencies and complex agent interactions, while Transformer-based architectures often suffer from high computational complexity and limited efficiency. To overcome these challenges, this paper proposes an efficient Mamba-based feature extraction framework for jointly encoding vehicle trajectories and map information. By leveraging state-space modeling and a selective scanning mechanism, the proposed approach effectively captures longrange dependencies and enhances the representation of complex traffic behaviors. Specifically, raw scene data are first normalized and embedded into a unified feature space. A Mamba Encoder is then employed to extract high-level features from historical vehicle trajectories and map elements. Subsequently, Vehicle-Vehicle and Vehicle-Map interaction modules are introduced to explicitly model dynamic interactions among traffic participants and between vehicles and the surrounding map. The resulting high-dimensional features are further fused using an additional Mamba Encoder, while a Global Interaction Module is designed to capture scenelevel dependencies. Finally, a Gated Recurrent Unit (GRU) decoder generates multi-modal future trajectory predictions. Experimental results on the Argoverse 1 dataset demonstrate that the proposed method achieves superior performance in terms of minADE, minFDE, and minMR, while maintaining high computational efficiency.OPEN ACCESS Received: 28/01/2026 Accepted: 16/04/2026

Abstract

Traditional methods for detecting surface defects in steel typically rely on manual visual inspection, eddy current testing, magnetic particle inspection, and machine vision. These methods often struggle to adapt to defects of varying scales and complex shapes, resulting in insufficient feature extraction, false detections, missed detections, and low detection accuracy. To address these issues, we propose an enhanced YOLOv8n model named MAM-YOLO, which aims to optimize defect detection on steel surfaces through multi-scale feature enhancement. By redesigning the convolution modules, the model improves accuracy in identifying surface defects, crazing, scratches, and patches. Furthermore, the design of a new module enhances processing capabilities for diverse surface features. A new loss function optimization is also applied to better identify crack characteristics. Experimental results on the NEU-DET dataset demonstrate that MAM-YOLO increases mAP by 5.4% and improves accuracy by 5.9% compared to the original YOLOv8n. Additionally, the model’s generalization capability was verified on the KolektorSDD dataset, achieving performance superior to the YOLOv8n baseline. Therefore, the proposed MAM-YOLO model offers significant advantages in precision and robust feature extraction for identifying steel surface defects.OPEN ACCESS Received: 25/11/2025 Accepted: 27/01/2026 Published: 21/07/2026