基于YOLOv8的卷烟物流中心姿态安全识别算法

打开文本图片集

中图分类号：TP391.41；F724.7 文献标识码：A

Abstract： Safety accidents in cigarette logistics centers are often caused by hazardous employee behaviors，while manual supervision suffers from low coverage and inefficiency. To address this issue，a posture safety recognition algorithm based on YOLOv8 is proposed for real-time monitoring，identification，and alarming of employee safety postures. A multi-scale attention mechanism module is integrated to enhance the model's perception and focus on multi-scale key features. Furthermore，an improved salient object detection strategy is employed to strengthen the sensitivity to local details of personnel，and a soft non-maximum suppression algorithm is introduced to mitigate false suppresson caused by overlapping targets. Experimental results in realworld scenarios demonstrate that the proposed algorithm achieves significant improvements across multiple performance metrics. The method enables accurate recognition and real-time alerting of hazardous behaviors among employees in cigarette logistics environments， providing a reliable and practical solution for intelligent safety monitoring and production management.

Keywords： YOLOv8 algorithm;multi-scale attention mechanism； salient object detection algorithm;soft non-maximum suppression algorithm

烟草工业的生产与物流环节复杂多样，安全生产管理要求高。（剩余9144字）