基于YOLOv4的烟叶烟梗识别算法研究

Research on Tobacco Stem Recognition Algorithm Based on YOLOv4

针对烟叶分级系统中烟叶上料、下料中的烟梗识别定位问题,提出了一种新型烟梗检测算法:基于YOLOv4卷积神经网络,将其主干网络修改替换为Efficientdet,降低网络模型的体积并提高模型对烟梗类型的识别准确率,在此基础上引入深度可分离卷积,并在深度可分离卷积中引入宽度因子alpha变量,优化了系统参数,简化了网络结构。结果表明,此算法在单姿态烟梗识别中其mAP值均优于SSD、Centernet、Faster-rcnn、YOLOv4little以及YOLOv4这几种目标检测算法,在多姿态烟梗识别中其mAP值和YOLOv4算法相差仅为2%,可减少网络参数,提高模型检测准确率与系统识别速度。

In response to the problem of identifying and locating tobacco stems during tobacco loading and unloading in tobacco grading systems,this paper proposed a YOLOv4 convolutional neural network.Based on this,depth separable convolution was intro‐duced,and the width factor alpha variable was introduced in depth separable convolution to optimize system parameters and simpli‐fied network structure.The experimental results showed that the mAP value of the algorithm proposed in this paper was superior to several object detection algorithms such as SSD,Centernet,Faster-rcnn,YOLOv4little,and YOLOv4 in single pose tobacco stem rec‐ognition.In multi pose tobacco stem recognition,the difference in mAP value between the algorithm and YOLOv4 algorithm is only 2%,but it can significantly reduce network parameters,improve model detection accuracy and system recognition speed.