检索规则说明:AND代表“并且”;OR代表“或者”;NOT代表“不包含”;(注意必须大写,运算符两边需空一格)
检 索 范 例 :范例一: (K=图书馆学 OR K=情报学) AND A=范并思 范例二:J=计算机应用与软件 AND (U=C++ OR U=Basic) NOT M=Visual
作 者:尹潇伟 孙仁诚 王霄鹏 邵峰晶 王光波 YIN Xiao-wei;SUN Ren-cheng;WANG Xiao-peng;SHAO Feng-jing;WANG Guang-bo(School of Computer Science and Technology,Qingdao University,Qingdao 266071,China)
机构地区:[1]青岛大学计算机科学技术学院,青岛266071
出 处:《青岛大学学报(自然科学版)》2022年第4期1-7,13,共8页Journal of Qingdao University(Natural Science Edition)
基 金:国家自然科学基金(批准号:41706198)资助。
摘 要:针对票据在识别时出现数据漏检率高、识别精度低的问题,提出文本检测模型ENCRAFT与识别模型DLCNN。在文本检测模型CRAFT的基础上,ENCRAFT修改其原始的特征提取网络的结构,利用未经池化的特征图进行融合,减少了细小特征的丢失,并增大监督特征图的分辨率,以提供更丰富的监督信息,从而提高模型检测率;DLCNN利用深层的卷积网络与浅层的循环网络实现对中文票据的高精度识别。实验结果表明,该方法在多个票据数据上的检测率与识别精度均有明显提升。There are problems of high data leakage rate and low recognition accuracy in the recognition of bills.Aiming at these problems,the text detection model ENCRAFT and the recognition model DLCNN were proposed.Based on the text detection model CRAFT,ENCRAFT is used to modify the structure of its original feature extraction network,and fuse the feature map without pooling.In the DLCNN model,the loss of fine features is reduced,and the resolution of the supervisory feature map is increased to provide richer supervisory information,so as to improve the model detection rate.A deep convolutional network and a shallow recurrent network were atilized to achieve high-precision identification of Chinese tickets.Experimental results show that the detection rate and recognition accuracy of the proposed method on multiple ticket data are significantly improved.
分 类 号:P391.41[天文地球—地球物理学]
正在载入数据...
正在载入数据...
正在载入数据...
正在载入数据...
正在载入数据...
正在载入数据...
正在载入数据...
正在链接到云南高校图书馆文献保障联盟下载...
云南高校图书馆联盟文献共享服务平台 版权所有©
您的IP:216.73.216.90