基于强化金字塔结构的P-ENet发票文本检测算法研究

打开文本图片集

中图分类号：TP391.4 文献标识码：A 文章编号：2096-4706（2025）11-0102-05

Research on P-ENet Invoice Text Detection Algorithm Based on Enhanced Pyramid Structure

LIUWenjie，LI Zhongwen，WANGYunfeng，ZHULi （ArmyLogisticsAcademy，Chongqing 40oo31，China）

Abstract：Asanindispensable document incommercial economicactivities，invoicesarecrucial to financialmanagement. However，the traditional invoicereimbursement processiscumbersomeand inefficient.Inrecent years，with therapid developmentofthe national economy，the frequencyofuseofinvoices has increased dramatically，making theautomatic procesing of invoices oneof the key means to improve the effciencyofbusiness operations.Therefore，aiming at the feldof invoice text detection，this paperconducts scientificresearch on the problems oflowaccuracyand slowrunning speedof the current popular text detectionalgorithms.BasedontheFPNarchitecture，this paper usesEficientNet and ASPP modules to constructanewinvoicetextdetectionnetworkP-ENetwithenhancedpyramidstructure.Thenexperimentsarecariedoutonthe self-annotated invoicedata set.The results show that the improved model hasan Intersection over Union of 86.89% ，aPrecision of 91.36% ，and a Recall of 85.43% on the invoice data set.Both have achieved a high level， and the running speed of the model has also been significantly improved.

Keywords：text detection; invoice recognition; EfficientNet; spatial pyramid

0 引言

发票在日常生活和工作中都扮演着很重要的角色，尤其是在企事业单位的结算报销业务中，发票被大量使用。（剩余6807字）