Haonan Yang

Real-time traffic object detection is essential for autonomous driving and intelligent transportation systems, where detection accuracy must be achieved under strict computational constraints. This study proposes a lightweight optimization framework for YOLOv5s-based traffic object detection and evaluates its effectiveness on a subset of the BDD100K benchmark. The framework incorporates three complementary optimization strategies. First, the original CIoU regression loss is replaced with SIoU to introduce directional information into bounding-box optimization and improve localization efficiency. Second, a Convolutional Block Attention Module (CBAM) is incorporated between the SPPF module and the neck to strengthen the representation of informative channel and spatial features in complex traffic scenes. Third, the genetic-algorithm-based hyperparameter evolution mechanism of YOLOv5 is employed to automatically adapt key training and augmentation parameters to the target traffic dataset. Experimental results show that the proposed method increases mAP@0.5 from 46.7% to 54.7% and mAP@0.5:0.95 from 23.2% to 27.1%, corresponding to absolute improvements of 8.0 and 3.9 percentage points, respectively. Class-level evaluation indicates particularly notable gains for traffic signs (+12.1 percentage points), traffic lights (+10.8 percentage points), and riders (+10.3 percentage points). Compared with larger YOLOv5 variants and YOLOv8s evaluated under the same protocol, the proposed detector provides competitive detection performance while retaining the compact characteristics of the YOLOv5s framework. The results demonstrate that coordinated optimization of the regression objective, feature representation, and training configuration can effectively improve lightweight traffic object detection without relying primarily on major increases in model complexity.

Automated vision-based structural health monitoring (SHM) of catenary support components is critical for ensuring railway operational safety. However, achieving reliable structural damage identification in practical engineering scenarios is hindered by complex environmental and operational variations (EOVs) in aerial imagery, the multi-scale nature of structural degradation, and the scarcity of damage samples. To address these SHM challenges, this paper proposes a high-precision unsupervised damage detection framework named Mask-Guided Asymmetrical Flow (MGAF). First, to mitigate the impact of EOVs, a SAM-guided preprocessing strategy is introduced to explicitly suppress background interference and extract effective structural Regions of Interest (ROIs). Second, an Asymmetrical Parallel Flow architecture is designed to balance damage detection sensitivity and processing latency. By optimizing the flow depth for high-resolution features, this architecture prevents overfitting to high-frequency environmental noise while preserving deep semantic information. Furthermore, a Cross-Scale Feature Fusion (CSFF) module is developed to enhance the detection capability for early-stage structural damages (e.g., fatigue micro-cracks and fastener looseness) by integrating global structural semantics with local textural damage indicators. Finally, a Morphological Edge Suppression (MES) mechanism is employed to eliminate boundary artifacts, thereby reducing the false alarm rate in pixel-level damage localization. Extensive experiments on the self-constructed CSCUD dataset demonstrate that the proposed system achieves competitive anomaly detection performance compared with representative unsupervised methods with an Image-level AUROC of 98.9% and a Pixel-level AP of 40.1%. During online inference, MGAF achieves a processing time of 0.077 s per frame on an NVIDIA RTX 4090 GPU, excluding the offline GSI-Net localization and SAM-based structural ROI extraction stages. This demonstrates the efficiency of the proposed anomaly detection module for practical railway inspection scenarios. Additional experiments on selected categories from the MVTec AD benchmark provide preliminary evidence of the transferability of MGAF to visually similar industrial anomaly detection scenarios.

Anomaly detection of catenary support components (CSCs) is an important component in railway condition monitoring systems. However, because the abnormal features of CSCs loosening are not obvious, and the current CNN models and visual Transformer models have problems such as limited remote modeling capabilities and secondary computational complexity, it is difficult for existing deep learning anomaly detection methods to effectively exert their performance. The state space model (SSM) represented by Mamba is not only good at long-range modeling, but also maintains linear computational complexity. In this paper, using the state space model (SSM), we proposed a new visual state space reconstruction network (VSM-UNet) for the detection of CSC loosening anomalies. First, based on the structure of UNet, a visual state space block (VSS block) is introduced to capture extensive contextual information and multi-scale features, and an asymmetric encoder–decoder structure is constructed through patch merging operations and patch expanding operations. Secondly, the CBAM attention mechanism is introduced between the encoder–decoder structure to enhance the model’s ability to focus on key abnormal features. Finally, a stable abnormality score calculation module is designed using MLP to evaluate the degree of abnormality of components. The experiment shows that the VSM-UNet model, learning strategy and anomaly score calculation method proposed in this article are effective and reasonable, and have certain advantages. Specifically, the proposed method framework can achieve an AUROC of 0.986 and an FPS of 26.56 in the anomaly detection task of looseness on positioning clamp nuts, U-shaped hoop nuts, and cotton pins. Therefore, the method proposed in this article can be effectively applied to the detection of CSCs abnormalities.