中圖分類號：TP391.41 文獻標(biāo)志碼：A DOI: 10.16157/j.issn.0258-7998.245589 中文引用格式： 王麗黎，樊盼盼，張詩雨. 面向密集行人場景的YOLOv8n改進算法[J]. 電子技術(shù)應(yīng)用，2025，51(2)：15-20. 英文引用格式： Wang Lili，F(xiàn)an Panpan，Zhang Shiyu. An improved YOLOv8n algorithm for dense pedestrian scenarios[J]. Application of Electronic Technique，2025，51(2)：15-20.
An improved YOLOv8n algorithm for dense pedestrian scenarios
Wang Lili1，2，F(xiàn)an Panpan1，Zhang Shiyu1
1.School of Automation and Information Engineering， Xi’an University of Technology； 2.Key Laboratory of Wireless Optical Communication and Network Research
Abstract： To address the issues of insufficient recognition accuracy and inaccurate detection of traditional algorithms in dense pedestrian scenarios, an improved dense pedestrian detection model based on YOLOv8n is proposed. Firstly, by introducing the SPPELAN module to replace the SPPF module in the backbone network, the model’s ability to perceive features of multi-scale targets is enhanced. Secondly, a residual attention mechanism is devised to improve the model’s ability to capture subtle features, thereby enhancing detection accuracy. Finally, by adding DySample operator and improving the small object detection layer, the model’s ability to locate and recognize small-scale objects is enhanced. Experimental results show that the improved model, compared to YOLOv8n, increases recall rate, mAP50, and mAP50-95 by 2.5%, 2.9%, and 2.4%, respectively, on the CrowdHuman dataset, and performs excellently on the WiderPerson and CityPersons datasets. The results of the experiments show that this algorithm is more effective for dense pedestrian detection tasks.
