摘要
为提高二维码在复杂环境下具有较好的检测速度和精度,提出了一种改进的YOLOv3-ms算法。通过增加SPP结构识别不同感受野特征和设置轻量特征金字塔模块融合高层语义特征与低层纹理特征,利用Mosaic增强方法扩充数据集和K-means算法重新聚类锚框提高YOLOv3-ms算法的检测精度。结果表明,与YOLOv3算法相比,召回率降低1.41%的情况下精确率提高了4%,推理速度提升了近3倍。
This paper proposes an improved YOLOv3-ms algorithm for the QR code to have a better detection speed and accuracy in a complex environment.The improvement is made possible by replacing the backbone feature extraction network in YOLOv3 with the lightweight network Mobilenetv2 and realizing the feature output of different receptive fields by adding the SPP structure block;applying the proposed lightweight feature pyramidmodule to integrate high level semantic features and low-level texture features;and ultimately expanding the data set with Mosaic Methods and re-clustering the anchor boxes using the K-means algorithm improving the detection accuracy of YOLOv3-ms algorithm.The results show that compared with the YOLOv3 algorithm,the improved algorithm enables an increased precision rate by 4%and a faster inference speed by nearly 3 times while achieving a recall rate reduction by 1.41%.
作者
杨庆江
臧佳琦
杨少辉
Yang Qingjiang;Zang Jiaqi;Yang Shaohui(School of Electronic & Information Engineering, Heilongjiang University of Science & Technology, Harbin 150022, China)
出处
《黑龙江科技大学学报》
CAS
2020年第6期692-697,共6页
Journal of Heilongjiang University of Science And Technology
