一种基于YOLOv5的二维码实时检测算法

Real-time detection algorithm of QR code based on YOLOv5

针对二维码图像检测过程中由于光线强度变化、图像环境复杂和拍摄设备移动等造成的识别精度低的问题,提出一种基于改进YOLOv5的目标检测方法来对二维码进行检测。将高效通道注意力机制(efficientchannelattention,ECA)加入YOLOv5的主干网络中,这种注意机制允许局部跨通道特征的相互作用而不需要降维,在节约算力的同时大大提升了网络的特征提取能力,有效地提高了检测精度。此外,将特征融合模块中原有特征金字塔结构替换成加权双向特征金字塔(BiFPN)结构,实现高效的加权特征融合和双向跨尺度连接,加快检测速度。对相应的二维码数据集进行了测试,结果表明,改进的YOLOv5模型平均准确率为97.6%,单张图片检测时间可达0.034s,相比于YOLOv5模型,平均精度提高了7.4%,达到了在复杂环境下对二维码目标精准与快速检测的要求。

Due to the low recognition accuracy brought on by changes in light intensity,a complex image environment,and the movement of shooting devices during QR code image detection,a target detection method based on improved YOLOv5 is proposed.The foundational network of YOLOv5 now includes a method for efficient channel attention(ECA).This attention method permits the interaction of local cross-channel information without dimensionality reduction,dramatically enhancing the network's ability to extract features and significantly enhancing detection accuracy while using less computational effort.To accomplish effective weighted feature fusion and bi-directional cross-scale connectivity for quicker detection,the original feature pyramid structure in the feature fusion module is replaced with a weighted bi-directional feature pyramid network(BiFPN)structure.The corresponding QR code dataset was put to the test,and the results show that the improved YOLOv5 model meets the criteria for accurate and quick detection of QR code targets in complex environments with mAP of 97.6%and a detection time of up to 0.034 s for a single image,an improvement in mAP of 7.4 percentage points over the YOLOv5 model.