| With the rapid development of China's logistics industry,the use of bills in logistics transportation is growing rapidly.Information in all bills needs to be stored in computer,which makes the pressure of processing bills more and more heavy.At present,there are two ways to extract information.The first way is by manual methods.This way requires the employment and training of professionals.It is time-consuming and labor-consuming,and the cost is high.And errors inevitably occur in work.The second is the automatic identification of simple forms of bills,which can only identify some simple and fixed forms of bills,such as bank checks.However,the bills used in the logistics field are generally tabular bills with complex formats,which facilitate input,storage and management information,and become a general form of bills in the logistics field.For this type of bills,there is no reliable technology that can quickly and accurately extract the key information in the logistics bills.Based on the above analysis,this paper makes a deep research on Logistics bills,and proposes an automatic key information extraction algorithm for complex forms of tabular bills,which can achieve rapid key information extraction of bills,with an accuracy of more than 98%.Firstly,this paper extracts the features of bills,and then automatically annotates the bills by clustering analysis,builds the data sets of bills,trains the classifier with these data sets,defines a template for each kind of bills,locates the location of the extracting unit by template matching,and obtains the key information in the extracting unit by text recognition.The work of this paper is as follows:1.Converting bills of different formats into image formats.Through the analysis of bills,it is found that bills form is the key feature of bills.Using digital image processing technology to process images,the information of bills form is obtained,and 23 feature points are extracted from these tables.2.Because the logistics bills have no class labeling,and the number of bills is large,we use clustering method to automatically calibrate the bills and construct training sets.Through the comparative analysis of various clustering algorithms,we choose K-Means++ as the clustering algorithm in this paper,and construct the data sets according to the clustering results.After comparing the performance of various classifiers,SVM is selected as the classifier of bill classification in this paper.3.In order to locate the extracting unit of bill quickly,this paper proposes a method of locating the extracting unit based on template matching.In addition,an interactive template definition tool is designed by PyQt,which can define template conveniently and quickly.4.Character recognition and similarity calculation are needed in bill template matching.In character recognition part,PDFMiner is used for plain text bills and optical character recognition technology is used for picture bills.In this paper,Levenshtein Distance is used to calculate character similarity. |
