中圖分類號：TP391.4 文獻標(biāo)志碼：A DOI: 10.16157/j.issn.0258-7998.256932 中文引用格式： 柯威曳，蘇吉才，齊騰濤，等. 一種基于YOLOv8模型的高速公路異常事件智能分析系統(tǒng)研究[J]. 電子技術(shù)應(yīng)用，2026，52(2)：39-44. 英文引用格式： Ke Weiye，Su Jicai，Qi Tengtao，et al. Research on an intelligent analysis system for highway abnormal events based on the YOLOv8 model[J]. Application of Electronic Technique，2026，52(2)：39-44.
Research on an intelligent analysis system for highway abnormal events based on the YOLOv8 model
Abstract： Event detection systems based on video analysis have been successful in many fields, but there are still issues such as missed detections, false positives, and insufficient accuracy in the field of highway anomaly detection. To address these issues, this paper improves YOLOv8 and uses artificial intelligence video analysis technology to propose a new Highway-YOLOv8 model and construct an intelligent video analysis system that can be used for highway anomaly detection. First, this paper designs a novel Deep Channel-by-Space Attention Mechanism (DCSM), which leverages channel and spatial interaction information to effectively enhance the model's field-of-view perception capabilities. Second, to address the loss of small object information in deep convolutional layers, this paper introduces a Multi-Stage Fusion Mechanism (MSFM), which significantly improves the model's small object detection capabilities. Finally, this paper adopts the advanced Wise-IoU loss function to replace the original YOLOv8 loss function, which effectively accelerates model convergence and improves detection accuracy. Experimental results show that Highway-YOLOv8 achieves a 2% improvement in mAP across all categories on the highway anomaly dataset compared to the original YOLOv8, with the highest improvement reaching 5% for single categories such as vehicles. This not only significantly enhances detection accuracy but also effectively reduces false negatives and false positives.
Key words : video analysis；YOLOv8；attention mechanism；multi-stage fusion
