摘要
在烟卷生产过程中,水渍、黄斑、褶皱和破损等缺陷是常见问题。传统的方法依赖于人工检测方法,但由于视觉疲劳等因素,导致生产效率不够高效。为了解决这一问题,本文提出了一种基于改进YOLOv8的烟卷污渍检测算法。本文以YOLOv8网络为骨干模型,加入了空间注意力机制,突出污渍区域的小目标检测能力,减小漏检和误检,从而减少小目标样本的误检率和漏检率。实验结果表明,与最新的YOLOv8算法相比,改进后的算法在烟卷污渍数据集上的平均精度均值(mAP)和召回率(Recall)分别提高了6.3%和1.8%;精度提高了6.2%。In the cigarette production process, defects such as water stains, yellow spots, wrinkles, and damage are common issues. Traditional methods rely on manual inspection, but due to factors like visual fatigue, production efficiency is not optimal. To address this problem, this paper proposes a cigarette stain detection algorithm based on an improved YOLOv8. We use the YOLOv8 network as the backbone model and incorporate a spatial attention mechanism to reduce the false detection and missed detection rates of small target samples. Experimental results show that, compared to the latest YOLOv8 algorithm, the improved algorithm increases the mean Average Precision (mAP) and Recall on the cigarette stain dataset by 6.3% and 1.8%, respectively, and improves precision by 6.2%.
出处
《图像与信号处理》
2024年第4期394-401,共8页
Journal of Image and Signal Processing
