复杂场景下的水表示数检测与识别  

Watermeter representation number detection and recognition in complex scenes

作　　者：康鑫 孙晓刚[1,2] 万磊 KANG Xin;SUN Xiaogang;WAN Lei(Chengdu Institute of Computer Application,Chinese Academy of Sciences,Chengdu Sichuan 610041,China;University of Chinese Academy of Science,Beijing 100049,China;Chengdu Zhongcai Xinda Technology Company Limited,Chengdu Sichuan 610041,China)

机构地区：[1]中国科学院成都计算机应用研究所,成都610041 [2]中国科学院大学,北京100049 [3]成都中钞信达科技有限公司,成都610041

出　　处：《计算机应用》2019年第S02期63-67,共5页journal of Computer Applications

基　　金：四川省重点研发计划项目(2018GZ0231)

摘　　要：目前自然场景下水表示数的统计工作由人工完成,但在大工作量的情况下人工统计容易出现统计错误和遗漏的情况,而且这一过程繁琐,耗时耗力。针对这种情况,提出一种基于深度学习的数字识别与检测算法。首先,使用旋转区域卷积神经网络(R2CNN)检测出文本框;然后,对该文本框区域使用基于注意机制(Attention)的文本识别算法预测出最终结果;最终,通过对不同深度的卷积神经网络(CNN)的速度和准确度进行对比,使用了一种准确有效的配置。结果显示训练好的网络模型可以应用于自然环境下的水表示数检测,能够达到实时检测的效果,并且优于其他检测识别算法的效果。不同数据下的结果验证了算法的有效性和鲁棒性。At present,the statistical work of watermeter number in natural scenes is done manually,but in the case of large workloads,manual statistics are prone to statistical errors and omissions,and this process is cumbersome,time consuming and labor intensive.Aiming at this situation,a digital recognition and detection algorithm based on deep learning was proposed.Firstly,a text box was detected using a Rotating Region Convolutional Neural Network(R2CNN);then a text recognition algorithm based on attention is used for the text box region to predict the final result.Finally,by comparing the speed and accuracy of different depth Convolutional Neural Networks(CNN),an accurate and efficient configuration was found.The results show that the trained network model can be applied to the detection of water representation in the natural environment,which can achieve real-time detection and is superior to other detection and recognition algorithms.The results under different data verify the effectiveness and robustness of the algorithm.

关 键 词：文本检测 文本识别 深度学习 旋转区域卷积神经网络 注意机制 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]