基于端到端深度网络的QR码快速检测算法

Fast QR Code Detection Algorithm Based on Deep Learning

在复杂环境下智能物流分拣系统中QR码检测任务有着较高的精度及速度要求,为满足工业现场硬件资源条件下算法模型的部署,提出了一种基于端到端网络的QR码快速检测算法。算法采用深度学习中主流One-Stage检测器作为基础框架,提取以及融合不同尺度的特征信息,同时在网络模型中加入优化的深度可分离卷积以及全连接注意力宏模块,以及应用快速非极大值抑制,整体上提升特征提取效果和模型检测的速度。最后,将文中算法与开源的经典模型(如YOLO V3、SSD-VGG、RetinaNet等)做对比实验,应用于采集于物流分拣现场QR码图片数据,结果表明上述算法在精度和速度两方面都有不错的表现,能够较好地解决实际工业应用问题。

The QR code detection task in the intelligent logistics sorting system in a complex environment has high accuracy and speed requirements.In order to meet the deployment of algorithm models under the conditions of industrial field hardware resources,we propose a fast QR code detection algorithm based on end-to-end network.The algorithm used the mainstream One-Stage detector in deep learning as the basic framework to extract feature information of different scales,and at the same time added a modified deep separable convolution and FCA module to the network model.Finally,fast non-maximum suppression was applied to improve the efficiency of feature extraction and the speed of model detection.Finally,a comparison experiment between the algorithm in this paper and the classic open-source algorithm(such as YOLO V3,SSD,RetinaNet,etc.)was applied to QR code image data at the logistics sorting site.The results show that the algorithm has good performance in both accuracy and speed,and can well solve practical industrial application problems.