pdf 文件版面分析--pdfplumber （python 文档解析提取）

pdfplumber 的特点

1、它是一个纯 python 第三方库，适合 python 3.x 版本
2、它用来查看pdf各类信息，能有效提取文本、表格
3、它不支持修改或生成pdf，也不支持对pdf扫描件的处理

import glob
import pdfplumber
import re
from collections import defaultdict
import json

class PDFProcessor:
    def __init__(self, filepath):
        self.filepath = filepath
        #打开文档，注意存放的位置
        self.pdf = pdfplumber.open(filepath)
        self.all_text = defaultdict(dict)
        self.allrow = 0
        self.last_num = 0

    def check_lines(self, page, top, buttom):
    	# 文本数据
        lines = page.extract_words()[::]
        text = ''
        last_top = 0
        last_check = 0
        for l in range(len(lines)):
            each_line = lines[l]
            check_re = '(?:。|；|单位：元|单位：万元|币种：人民币|\d|报告(?:全文)?(?:（修订版）|（修订稿）|（更正后）)?)$'
            if top == '' and buttom == '':
                if abs(last_top - each_line['top']) <= 2:
                    text = text + each_line['text']
                #elif last_check > 0 and (page.height * 0.85 - each_line['top']) > 0 and not re.search(check_re, text):
                elif last_check > 0 and (page.height * 0.9 - each_line['top']) > 0 and not re.search(check_re, text):

                    text = text + each_line['text']
                else:
                    text = text + '\n' + each_line['text']
            elif top == '':
                if each_line['top'] > buttom:
                    if abs(last_top - each_line['top']) <= 2:
                        text = text + each_line['text']
                    elif last_check > 0 and (page.height * 0.85 - each_line['top']) > 0 and not re.search(check_re,
                                                                                                          text):
                        text = text + each_line['text']
                    else:
                        text = text + '\n' + each_line['text']
            else:
                if each_line['top'] < top and each_line['top'] > buttom:
                    if abs(last_top - each_line['top']) <= 2:
                        text = text + each_line['text']
                    elif last_check > 0 and (page.height * 0.85 - each_line['top']) > 0 and not re.search(check_re,
                                                                                                          text):
                        text = text + each_line['text']
                    else:
                        text = text + '\n' + each_line['text']
            last_top = each_line['top']
            last_check = each_line['x1'] - page.width * 0.85

        return text

    def drop_empty_cols(self, data):
        # 删除所有列为空数据的列
        transposed_data = list(map(list, zip(*data)))
        filtered_data = [col for col in transposed_data if not all(cell is '' for cell in col)]
        result = list(map(list, zip(*filtered_data)))
        return result

    @staticmethod
    def keep_visible_lines(obj):
        """
        If the object is a ``rect`` type, keep it only if the lines are visible.

        A visible line is the one having ``non_stroking_color`` not null.
        """
        if obj['object_type'] == 'rect':
            if obj['non_stroking_color'] is None:
                return False
            if obj['width'] < 1 and obj['height'] < 1:
                return False
            # return obj['width'] >= 1 and obj['height'] >= 1 and obj['non_stroking_color'] is not None
        if obj['object_type'] == 'char':
            return obj['stroking_color'] is not None and obj['non_stroking_color'] is not None
        return True

    def extract_text_and_tables(self, page):
        buttom = 0
        page = page.filter(self.keep_visible_lines)
        tables = page.find_tables()
        if len(tables) >= 1:
        	# 表格数据
            count = len(tables)
            for table in tables:
                if table.bbox[3] < buttom:
                    pass
                else:
                    count -= 1
                    top = table.bbox[1]
                    text = self.check_lines(page, top, buttom)
                    text_list = text.split('\n')
                    for _t in range(len(text_list)):
                        self.all_text[self.allrow] = {'page': page.page_number, 'allrow': self.allrow,
                                                      'type': 'text', 'inside': text_list[_t]}
                        self.allrow += 1

                    buttom = table.bbox[3]
                    new_table = table.extract()
                    r_count = 0
                    for r in range(len(new_table)):
                        row = new_table[r]
                        if row[0] is None:
                            r_count += 1
                            for c in range(len(row)):
                                if row[c] is not None and row[c] not in ['', ' ']:
                                    if new_table[r - r_count][c] is None:
                                        new_table[r - r_count][c] = row[c]
                                    else:
                                        new_table[r - r_count][c] += row[c]
                                    new_table[r][c] = None
                        else:
                            r_count = 0

                    end_table = []
                    for row in new_table:
                        if row[0] != None:
                            cell_list = []
                            cell_check = False
                            for cell in row:
                                if cell != None:
                                    cell = cell.replace('\n', '')
                                else:
                                    cell = ''
                                if cell != '':
                                    cell_check = True
                                cell_list.append(cell)
                            if cell_check == True:
                                end_table.append(cell_list)
                    end_table = self.drop_empty_cols(end_table)

                    for row in end_table:
                        self.all_text[self.allrow] = {'page': page.page_number, 'allrow': self.allrow,
                                                      'type': 'excel', 'inside': str(row)}
                        # self.all_text[self.allrow] = {'page': page.page_number, 'allrow': self.allrow, 'type': 'excel',
                        #                               'inside': ' '.join(row)}
                        self.allrow += 1

                    if count == 0:
                        text = self.check_lines(page, '', buttom)
                        text_list = text.split('\n')
                        for _t in range(len(text_list)):
                            self.all_text[self.allrow] = {'page': page.page_number, 'allrow': self.allrow,
                                                          'type': 'text', 'inside': text_list[_t]}
                            self.allrow += 1

        else:
        	#文本数据
            text = self.check_lines(page, '', '')
            text_list = text.split('\n')
            for _t in range(len(text_list)):
                self.all_text[self.allrow] = {'page': page.page_number, 'allrow': self.allrow,
                                              'type': 'text', 'inside': text_list[_t]}
                self.allrow += 1

        first_re = '[^计](?:报告(?:全文)?(?:（修订版）|（修订稿）|（更正后）)?)$'
        end_re = '^(?:\d|\\|\/|第|共|页|-|_| ){1,}'
        if self.last_num == 0:
            try:
                first_text = str(self.all_text[1]['inside'])
                end_text = str(self.all_text[len(self.all_text) - 1]['inside'])
                if re.search(first_re, first_text) and not '[' in end_text:
                    self.all_text[1]['type'] = '页眉'
                    if re.search(end_re, end_text) and not '[' in end_text:
                        self.all_text[len(self.all_text) - 1]['type'] = '页脚'
            except:
                print(page.page_number)
        else:
            try:
                first_text = str(self.all_text[self.last_num + 2]['inside'])
                end_text = str(self.all_text[len(self.all_text) - 1]['inside'])
                if re.search(first_re, first_text) and '[' not in end_text:
                    self.all_text[self.last_num + 2]['type'] = '页眉'
                if re.search(end_re, end_text) and '[' not in end_text:
                    self.all_text[len(self.all_text) - 1]['type'] = '页脚'
            except:
                print(page.page_number)

        self.last_num = len(self.all_text) - 1


    def process_pdf(self):
        for i in range(len(self.pdf.pages)):
            self.extract_text_and_tables(self.pdf.pages[i])


    def save_all_text(self, path):
        with open(path, 'w', encoding='utf-8') as file:
            for key in self.all_text.keys():
                file.write(json.dumps(self.all_text[key], ensure_ascii=False) + '\n')


def process_all_pdfs_in_folder(folder_path):
    file_paths = glob.glob(f'{folder_path}/*')
    file_paths = sorted(file_paths, reverse=True)

    for file_path in file_paths:
        print(file_path)
        try:
            processor = PDFProcessor(file_path)
            processor.process_pdf()
            save_path = 'RAG_ASMPLE_DATAS_TXTS/' + file_path.split('/')[-1].replace('.pdf', '.txt')
            processor.save_all_text(save_path)
        except:
            print('check')

if __name__ == '__main__':
    # 需要解析的pdf文件路径
    pdf_path = r'C:\Users\WWS\RAG_ASMPLE_DATAS\2020-02-26__上海爱旭新能源股份有限公司__600732__爱旭股份__2019年__年度报告.pdf'
    # pdf解析后的txt内容文件
    out_path = r'C:\Users\WWS\RAG_ASMPLE_DATAS\2020-02-26__上海爱旭新能源股份有限公司__600732__爱旭股份__2019年__年度报告.txt'
    processor = PDFProcessor(pdf_path)
    processor.process_pdf()
    processor.save_all_text(out_path)

参考

版面分析–PDF解析神器pdfplumber
版面分析–富文本txt读取

补充

---

提取PDF中的图片并保存到本地

import pdfplumber
file_name = "**.pdf"# 需要解析的pdf的文件路径
output_file = "**.xlsx" # pdf解析后的内容

with pdfplumber.open(file_name) as pdf:
	#获取第一页
	first_page = pdf.pages[1]
	print('页码：', first_page.page_number)
	print('page width:', first_page.width)
	print('page height:', first_page.height)
	# get the first page text
	text = first_page.extract_text()
	print(text)
	
	# 获取第一页图片，获取到的是一个列表，列表中存储的是字典
	imgs = first_page.images
	i = 0
	for img in imgs:
		# 获取图片的二进制流
		print(img['stream'].get_data())
		with open(output_file, mode='wb') as f2:
			f2.write(img['stream'].get_data())

提取pdf 表格文本，保存为excel文件

import pdfplumber
from openpyxl import Workbook# 保存表格，需要安装openpyxl
file_name = '**.pdf'
output_file = '**.xlsx'
with pdfplumber.open(file_name) as pdf:
	page01 = pdf.pages[0]
	table = page01.extract_table()
	workbook = Workbook()
	sheet = workbook.active
	for row in table:
		sheet.append(row)
	workbook.save(filename=output_file)

提取PDF表格 文本

import pdfplumber
file_name = '**.pdf'
output_file = '**.txt'
with pdfplumber.open(file_name) as p:
	page_count = len(p.pages)# 统计文档的页数
	for i in range(0, page_count):
		page = p.pages[i]# 提取每页的对象并存储
		textdata = page.extract_table()#提取每页的表格文字信息
		# table2 = page01.extract_tables()# 提取多个表格
		data = open(output_file , 'a') # 将 表格文字存放在需要存储的文档里面
		data.write(textdata )# 文档内容写入

提取PDF纯文本

import pdfplumber
file_name = '**.pdf'
output_file = '**.txt'
with pdfplumber.open(file_name) as p:
	page_count = len(p.pages)# 统计文档的页数
	for i in range(0, page_count):
		page = p.pages[i]# 提取每页的对象并存储
		textdata = page.extract_text()#提取每页的文字信息
		data = open(output_file , 'a') # 将 表格文字存放在需要存储的文档里面
		data.write(textdata )# 文档内容写入
	

读取富文本txt
python 读取文件函数有三种 read()、readline()、readlines()

• read() 一次性读取所有文本
• readline() 读取第一行的内容
• readlines() 读取全部内容，以数列的格式返回

with open('rag_datas/story.txt', 'r', encoding='utf-8' ) as f:
	data = f.read()
	print(data)

with open('rag_datas/story.txt', 'r', encoding='utf-8' ) as f:
	data = f.readline()
	print(data)

with open('rag_datas/story.txt', 'r', encoding='utf-8' ) as f:
	for line in f.readlines():
		line = line.strip('\n')
		print(line)