面向精准价格牌识别的多任务循环神经网络

为了促进智能新零售在线下业务场景的发展, 提高作为销售关键信息价格牌的识别精度. 本文对价格牌识别问题进行研究, 有效地提高了价格牌的识别精度, 并解决小数点定位不准确的难题. 通过深度卷积神经网络提取价格牌的深度语义表达特征, 将提取到的特征图送入多任务循环网络层进行编码, 然后根据解码网络设计的注意力机制解码出价格数字, 最后将多个分支的结果整合并输出完整价格. 本文所提出的方法能够非常有效地提高线下零售场景价格牌的识别精度, 并解决了一些领域难题如小数点的定位问题, 此外, 为了验证本文方法的普适性, 在其他场景数据集上进行了对比实验, 相关结果也验证了本文方法的有效性.

Towards Accurate Price Tag Recognition Algorithm With Multi-task RNN

In order to promote the development of smart new retail in the offline scenario and improve the recognition accuracy of price tags,which is a key sales information,this paper studies this application scene to improve the recognition accuracy effectively of the price tag and solves the difficulty of locating decimal point.The deep semantic expression features of price tag are extracted by the deep convolutional neural network,sent to the multi-task recurrent network layer for encoding,and then the price number is decoded according to the attention mechanism of the decoding network,and finally the results of multiple branches are integrated to output the complete price.The method we proposed can effectively improve the price recognition accuracy in the smart new retail scenario and solve some challenge problems such as locating the position of decimal point.In addition,in order to verify the universality of the method,comparative experiments are carried out on datasets of other scenarios,and the related results also verify the effectiveness of the method in this paper.