基于深度学习的条烟分拣线上烟包错配识别系统的构建

  背景  条烟分拣线上，条烟长边相邻并排摆放形成一层，多层叠加形成烟包，与订单相比，烟包可能存在少烟、多烟、品规错误等问题，目前采用的人工检查方式效率较低，且难以完全避免错误发生。本研究的目的是构建烟包错配识别系统。  方法  采用由工业相机镜头和光源构成的机器视觉系统采集成品烟包侧面与顶面图像，以基于深度学习的物体定位和识别技术获取烟包中条烟的数量与品规，与物流上位系统订单数据比对，自动识别与提示错误烟包。  结果  （1）实际使用中烟包识别成功率≥99.99%，识别耗时≤300 ms。识别过程与原有工作步骤并行，增加识别系统不降低分拣效率。（2）系统上线运行至今有效避免了烟包连续出错和返工问题。（3）识别系统可以减轻搬运工人的工作负担，进而提高工作效率。  结论  采用深度学习机器视觉系统自动化识别烟包品规，可以提升烟草物流条烟分拣的质量和效率。 

Construction of cigarette pack mismatch recognition system based on deep learning on cigarette sorting line

  Background  In tobacco logistics, several cigarettes are placed side by side to form a layer, and and multiple layers are superimposed to form cigarette packs, which may lead to problems such as fewer cigarettes, more cigarettes, and incorrect cigarette specifications. At present, the manual inspection method is inefficient and difficult to avoid errors completely. Correcting the error after the delivery of the wrong cigarette pack is time-consuming and labor-intensive, and the failure to find the wrong cigarette pack in time during production will result in multiple consecutive errors, which need to be addressed by time-consuming rework.  Methods  Machine vision system based on industrial camera lens and light was used to collect the side and top surface images of the finished cigarette pack, and the object positioning and recognition technology based on deep learning was used to obtain the quantity and product specifications of the cigarettes in the cigarette pack, and compare them with the order data of logistics upper system to automatically identify and prompt wrong cigarette pack.  Results  (1) In actual use, the recognition success rate was ≥ 99.99%, and the time cost was ≤ 300 milliseconds. The recognition process was parallel to the original work steps, and the addition of the recognition system did not reduce the sorting efficiency. (2) After the system went online, the problem of continuous errors in recognizing and sorting cigarette packs was effectively avoided, and a large number of rework were avoided. (3) The identification system can reduce the workload of the operators, thereby improving work efficiency.  Conclusion  Deep learning machine vision system can improve the quality and efficiency of cigarette sorting in tobacco logistics..