Xiaoming Chen

Detecting unmanned aerial vehicles (UAVs) in complex low-altitude scenes remains challenging due to limited target details and background interference. To address these challenges, this paper proposes DARA-Net, a lightweight UAV detection framework that coordinates detail preservation, efficient feature extraction, and multiscale aggregation. Built upon YOLOv11, DARA-Net integrates adaptive window reduction (AWR) to retain informative local features during downsampling, a separable aggregation block (SAB) to improve feature representation with fewer parameters, and a directional weighted pyramid (DWP) to enhance cross-scale feature fusion. Experiments on a visible-light UAV dataset containing 25,941 images demonstrate that DARA-Net achieves 97.0% precision, 91.1% recall, 95.1% mAP@0.5, and 62.0% mAP@0.5:0.95. Compared with the YOLOv11 baseline, recall improves by 2.6 percentage points, while the parameter count decreases by approximately 30.6% and model size decreases by 29.1%, reaching 1.79 M parameters and 3.9 MB, respectively. Compared with YOLOv26n, DARA-Net achieves a higher mAP@0.5:0.95 (62.0% versus 60.77%) with approximately 24.6% fewer parameters and a 28.2% smaller model file, while maintaining comparable mAP@0.5. Additional validation on an external UAV dataset further supports its competitive detection performance. These results highlight the value of DARA-Net as a compact UAV detector that prioritizes localization performance and reduced parameter and storage requirements.

Accurate detection of distant drones requires a compact network that preserves weak visual cues. We combine a separable context pyramid (SCP), dual-context energy attention (DCEA), and geometry-calibrated regression (GCR) within a YOLO11n detector. Across repeated training runs on UAV-6E, the full model reached 94.25% validation mAP@0.5 (SD 0.29%), compared with 91.13% (SD 0.59%) for YOLO11n and 93.58% (SD 0.21%) for SCP alone. SCP supplied most of the gain. In the external DUT-2200 evaluation, full-model AP@0.75 was 54.55% (SD 0.47%) versus 51.13% (SD 1.80%) for YOLO11n. The full detector used 2.33 million parameters and 5.27 GFLOPs at 640 by 640 pixel input, compared with 2.59 million and 6.44 GFLOPs for YOLO11n. The embedded implementation achieved 26.3 FPS on the Orange Pi 5 Plus RK3588 platform, supporting practical camera-based UAV monitoring.

In the context of rapid advancements in the automotive industry and intelligent transportation systems, assessing cognitive load during driving has become a key factor for driving safety and user experience. This paper presents a comprehensive narrative review of theories, methods, and technological advancements in assessing cognitive load during driving. Rather than following a systematic review protocol,the structure of the review is organized around key research questions and critical gaps identified in the current literature. We assess the applicability and performance of major evaluation methods, including physiological indicators, behavioral measures, subjective self-report scales, and data-driven approaches, across various driving scenarios. The review also discusses the integration of multi-source information and propose a conceptual framework for holistic and adaptive cognitive load assessment. Furthermore, it highlights challenges in current practices, such as technical constraints, environmental variability, and individual differences. Special emphasis is placed on the relevance of cognitive load assessment for industrial informatics, particularly in the context of advanced driver assistance systems (ADAS), autonomous driving technologies, and driver training programs. This review aims to provide a structured synthesis of current approaches, offer practical insights for application and system design, and guide future research toward developing more robust, generalizable, and context-aware assessment tools. By analyzing the state of the art, we contribute a timely reference to support both academic development and industrial implementation for next-generation intelligent vehicle systems.

Functional near-infrared spectroscopy (fNIRS) is widely used to monitor cognitive load, yet classification in dynamic, cooperative settings remains challenging due to noise and hemodynamic delays. This article aims to develop a biologically inspired classification framework for fNIRS data that is suitable for both individual and collaborative learning environments. We propose fNIRS-SpikeNet, a spiking neural network (SNN) framework that integrates rate, latency, and delta spike encoding strategies with a residual-SNN architecture to capture spatiotem-poral dynamics. We evaluate our method on three public fNIRS datasets involving mental and motor tasks. Experimental results demonstrate that fNIRS-SpikeNet, particularly under rate encoding, significantly outperforms conventional machine learning and deep learning baselines in accuracy, efficiency, and real-time adaptability. These outcomes highlight the potential of SNNs for low-power, real-time neuroimaging in socially interactive applications.