一种改进YOLOv7的瓦楞纸箱表面缺陷检测方法

Surface defect detection method for corrugated carton based on improved YOLOv7

针对物流过程中对瓦楞纸箱表面外观检测,通用的检测算法在面对小目标、低光照、背景复杂的情况下,不能有效地对缺陷目标进行检测。为此,本文提出一种基于改进YOLOv7的瓦楞纸箱表面缺陷检测方法,有效地提升检测效果。首先,在YOLOv7的骨干网络中加入AIFI模块,通过有效地提取特征来提升模型的检测精度,还可以减少计算冗余。同时,为了进一步提高模型检测性能,添加感受野注意力卷积运算(RFAConv)模块,利用空间注意机制关注感受场的空间特征,与卷积的结合消除了卷积参数共享的问题,再通过全局平均池化来获取全局信息,再次提升模型的检测性能。实验结果证明,本文算法相较于原YOLOv7算法,在模型大小、参数量基本保持不变的情况下,检测精度mAP从原来的96.3%提高到了97.5%,总体提升了1.2%,有效验证了本文改进算法的有效性。

In response to the appearance of the surface appearance of the corrugated carton in the actual logistics process,the general detection algorithm cannot effectively detect the defective targets in the case of small targets,low light,and complex backgrounds.To this end,this article proposes a corrugated carton surface defect detection method based on improving YOLOv7 to effectively improve the detection effect.This article first adds the AIFI module to the backbone network of YOLOv7,through effective extraction features,the detection accuracy of the model can be improved,and the redundancy can be reduced.At the same time,in order to further im-prove the model detection performance,add the RFAConv module,use the space to pay attention to the spatial characteristics of the feeling field,combine with the convolution to eliminate the problem of convolution parameters sharing,and then obtain global informa-tion through a global average pooling,and improve the model detection performance again.The experimental results prove that com-pared with the original YOLOv7 algorithm,the algorithm of this article has increased from 96.3%to 97.5%,overall increased by 1.2%,effectively verify the effectiveness of the algorithm in this article.