Yanfeng Lu

Efficient object detection is needed in automated driving and edge perception. In these scenarios, a detector must work under limits on latency, power, and memory. YOLOv8 is a strong real-time baseline, but its computation can still be high for compact deployment. This paper proposes MV3-YOLO, a lightweight YOLOv8 variant with a stage-wise hybrid backbone. The early Conv/C2f stages are kept to retain low-level spatial details. Lightweight modules are placed in deeper stages, where feature maps are smaller and redundant computation is more common. C2fMixed is used at the stride-16 stage to balance feature capacity and cost. C2fGhostis used at the deepest stage to generate high-level features with fewer parameters. The YOLOv8 neck and head are kept unchanged for stable multi-scale fusion. On the KITTI validation set, MV3-YOLO reaches mAP@0.5 = 0.859 and mAP@0.5:0.95 = 0.610 with only 2.53 M parameters and 6.6 GFLOPs. Compared with YOLOv8n, it reduces parameters by 19.7% and GFLOPs by 25.0% while improving mAP@0.5 by 1.66% and mAP@0.5:0.95 by 1.50%. On COCO val2017, MV3-YOLO obtains 38.4 mAP@0.5:0.95, which is higher than the YOLOv8n reference result and close to YOLOv10n. These results show that MV3-YOLO reduces deployment cost while keeping competitive detection accuracy.

Spiking Neural Networks (SNNs), with their brain-inspired spatiotemporal dynamics and spike-driven computation, have emerged as promising energy-efficient alternatives to Artificial Neural Networks (ANNs). However, existing SNNs typically replicate inputs directly or aggregate them into frames at fixed intervals. Such strategies lead to neurons receiving nearly identical stimuli across time steps, severely limiting the model's expressive power, particularly in complex tasks like object detection. In this work, we propose the Temporal Dynamics Enhancer (TDE) to strengthen SNNs' capacity for temporal information modeling. TDE consists of two modules: a Spiking Encoder (SE) that generates diverse input stimuli across time steps, and an Attention Gating Module (AGM) that guides the SE generation based on inter-temporal dependencies. Moreover, to eliminate the high-energy multiplication operations introduced by the AGM, we propose a Spike-Driven Attention (SDA) to reduce attention-related energy consumption. Extensive experiments demonstrate that TDE can be seamlessly integrated into existing SNN-based detectors and consistently outperforms state-of-the-art methods, achieving mAP50-95 scores of 57.7% on the static PASCAL VOC dataset and 47.6% on the neuromorphic EvDET200K dataset. In terms of energy consumption, the SDA consumes only 0.240 times the energy of conventional attention modules.

Diffusion models have shown significant potential in imitation learning, particularly for modeling complex data distributions. However, challenges remain in effectively capturing temporal dependencies in sequential decision-making tasks that are crucial for accurately representing the dynamics of actions and predicting long-term outcomes. This article proposes temporally masked diffusion (TM-Diff), an innovative approach that enhances the learning of essential temporal dependencies within action sequences. The proposed TM-Diff introduces a temporal masking method, masking certain actions at the time-step level during training. A novel time-series diffusion Transformer is designed to reconstruct the masked positions as a self-supervised learning (SSL) auxiliary task, while still following the diffusion training process. This enables TM-Diff to infer missing action information, allowing it to learn temporal relationships between action sequence tokens. The effectiveness of TM-Diff is validated across six complex manipulation tasks. The experimental results indicate that TM-Diff demonstrates improved generative performance compared to standard diffusion-based policies and traditional observation-to-action methods. Particularly in small-sample demonstration datasets, TM-Diff achieves notable success, with a 6.2% improvement in the success rate for the square task and a 10.1% improvement in the target coverage ratio for the Push-T task, using only 50 demonstration samples. As a reliable and robust task execution policy, the proposed TM-Diff supports the responsible deployment of intelligent robotic systems in critical domains such as the Industrial Internet of Things (IIoT), home services, healthcare, and logistics.

This study presented an autonomous shield-cutting end-effector for maize surrounding weeding (SEMSW), addressing the challenges of the low weed removal rate (WRR) and high seedling damage rate (SDR) in northern China’s 3–5 leaf stage maize. The SEMSW integrated seedling positioning, robotic arm control, and precision weeding functionalities: a seedling positioning sensor identified maize seedlings and weeds, guiding XYZ translational motions to align the robotic arm. The seedling-shielding anti-cutting mechanism (SAM) enclosed crop stems, while the contour-adaptive weeding mechanism (CWM) activated two-stage retractable blades (TRWBs) for inter/intra-row weeding operations. The following key design parameters were determined: 150 mm inner diameter for the seedling-shielding disc; 30 mm minimum inscribed-circle for retractable clamping units (RCUs); 40 mm ground clearance for SAM; 170 mm shielding height; and 100 mm minimum inscribed-circle diameter for the TRWB. Mathematical optimization defined the shape-following weeding cam (SWC) contour and TRWB dimensional chain. Kinematic/dynamic models were introduced alongside an adaptive sliding mode controller, ensuring lateral translation error convergence. A YOLOv8 model achieved 0.951 precision, 0.95 mAP50, and 0.819 mAP50-95, striking a balance between detection accuracy and localization precision. Field trials of the prototype showed 88.3% WRR and 2.2% SDR, meeting northern China’s agronomic standards.

Visual target detection under adverse weather conditions presents a fundamental challenge for autonomous driving, particularly in achieving all-weather operational capabilities. Unlike existing reviews that concentrate on individual technical domains such as image restoration or detection robustness, this review introduces an innovative “restoration–detection” collaborative framework. This paper systematically examines state-of-the-art methods for degraded image recovery and improvement of detection model robustness, encompassing from traditional, physically driven approaches as well as contemporary deep learning paradigms. A comprehensive overview and comparative analysis are provided to elucidate these advancements. Regarding the recovery of degraded images, traditional methods demonstrate advantages in interpretability within specific scenarios, such as those based on dark channel prior. In contrast, deep learning methods have achieved significant breakthroughs in modeling complex degradations and enhancing cross-domain generalization through a data-driven paradigm. In the field of enhancing detection robustness, traditional improvement techniques that utilize anisotropic filtering, alongside deep learning methods such as SSD, R-CNN, and the YOLO series, contribute to perceptual stability through feature optimization and end-to-end learning approaches, respectively. This paper summarizes 11 types of mainstream public datasets, examining their multimodal annotation system and addressing issues related to discrepancies. Furthermore, it provides an extensive evaluation of algorithm performance using PSNR, SSIM, mAP, among others. It has been identified that significant bottlenecks persist in dynamic weather coupling modeling, multimodal heterogeneous data fusion, and the efficiency of edge deployment. Future research should focus on establishing a physically guided hybrid learning architecture, developing techniques for dynamic and adaptive timing calibration, and designing a flexible multimodal fusion framework to overcome the limitations associated with complex environment perception. This paper serves as a systematic reference for both the theoretical development and practical implementation of automatic driving vision detection technology under severe weather conditions.

The accuracy of object detection is crucial for the safety and efficiency of traffic management in monitoring systems. Existing detectors, however, struggle significantly within complex urban scenarios where high-density occlusions among the targets occur, as well as extreme scale variations resulting from the size differences of vehicles and distance variations to the camera. To remedy these issues, we introduce DFE-YOLO, an enhanced multi-scale detection framework built upon YOLOv8 that fuses features from various layers at different scales through our ‘four adaptive spatial feature fusion’ module, which performs adaptive spatial fusion via learnable weights normalized by softmax and thereby allows effective feature aggregation across scales. The second contribution is DySample, which uses a lightweight, content-aware, point-based upsampling method to improve multi-scale feature representation as well as reduce imbalance across different object scales. The experiments conducted on the VisDrone-2019 and BDD100K benchmarks showed significantly superior performance against state-of-the-art detectors. Specifically, DFE-YOLO achieved a +4% and +5.1% boost over YOLOv10 in AP and APsmall. This study offers a useful fix for smart transport systems.