基于改进 YOLO v4 的轻量化烟梗识别方法

为完成烟叶精选工艺流程中打叶复烤后破碎烟叶的进一步去梗,实现破碎烟叶中烟梗的自动化检测,文章提出基于改进YOLO v4的轻量化烟梗识别方法。在YOLO v4基础模型上先后进行通道剪枝和层剪枝,大幅简化模型结构,改进后模型存储空间下降了93.77%,模型平均精度均值(mean average precision, mAP)和前向运算时间与基础模型持平。与同类别算法相比,模型精度平均提升8.7%,模型参数量大幅缩减。实验结果表明该实验剪枝模型更具轻量化,识别效果更好,能够满足实际生产需求。

In order to realize further stem removal of broken tobacco leaves after leaf beating and re-roasting in tobacco leaf selection process and realize automatic detection of tobacco stems in broken tobacco leaves, a lightweight tobacco stem identification method based on improved YOLO v4 was proposed. In YOLO V4 model, channel pruning and layer pruning were carried out successively to simplify the model structure greatly. The storage space of the improved model is reduced by 93.77%, and the mean average precision (mAP) of the model and the forward computing time are the same as those of the basic model. Compared with algorithms of the same category, the model accuracy is improved by 8.7% on average, and the number of model parameters is greatly reduced. The analysis shows that the pruning model is more lightweight, has better identification effect, and meets the actual production demand.