pip install pandas camelot-py pdfplumber pywin32

import pandas as pd
import camelot  # 基于文本的 PDF，而不適用于掃描文檔
import os
from win32com.client import Dispatch
from os import walk
import pdfplumber
# 文件夾路徑
path = 'E:\\data\\table_json-master'
# 閾值，判斷表格是否位于頁面頂部或底部
topthreshold = 0.2
dthreshold = 0.8
# 將word文件轉(zhuǎn)化為PDF
wdFormatPDF = 17

def doc2pdf(input_file):
    word = Dispatch('Word.Application')
    doc = word.Documents.Open(input_file)
    doc.SaveAs(input_file.replace(".docx", ".pdf"), FileFormat=wdFormatPDF)
    doc.Close()
    word.Quit()


# 遍歷文件夾，將其中的word文件轉(zhuǎn)化為PDF
for root, dirs, filenames in walk(path):
    for file in filenames:
        if file.endswith(".doc") or file.endswith(".docx"):
            doc2pdf(os.path.join(root, file))

def is_first_table_at_top_of_page(page):
    page_height = page.height
    table_bounds_list = get_table_bounds(page)
    if table_bounds_list:
        first_table_bounds = table_bounds_list[0]
        is_first_table_at_top = first_table_bounds[1] < topthreshold * page_height
        return {
            "is_first_table_at_top": is_first_table_at_top,
        }
    else:
        return {
            "is_first_table_at_top": False,
        }


def is_last_table_at_bottom_of_page(page):
    page_height = page.height
    table_bounds_list = get_table_bounds(page)
    if table_bounds_list:
        last_table_bounds = table_bounds_list[-1]
        is_last_table_at_bottom = last_table_bounds[3] > dthreshold * page_height
        return {
            "is_last_table_at_bottom": is_last_table_at_bottom,
        }
    else:
        return {
            "is_last_table_at_bottom": False,
        }


def get_table_bounds(page):
    table_bounds_list = []
    tables = page.find_tables()
    for table in tables:
        x0, top, x1, bottom = table.bbox
        table_bounds_list.append((x0, top, x1, bottom))
    return table_bounds_list


def is_table_ending_on_page(pdf_path, page, npage):
    tables = camelot.read_pdf(pdf_path, pages=str(npage + 1))
    if not tables:
        return False, 0
    table = tables[-1]
    last_row = table.df.iloc[-1]
    last_table_result = is_last_table_at_bottom_of_page(page)
    return last_table_result['is_last_table_at_bottom'], len(last_row)


def is_table_starting_on_page(pdf_path, page, npage):
    tables = camelot.read_pdf(pdf_path, pages=str(npage + 1))
    if not tables:
        return False, 0
    table = tables[0]
    first_row = table.df.iloc[0]
    first_table_result = is_first_table_at_top_of_page(page)
    return first_table_result['is_first_table_at_top'], len(first_row)

# 獲取文件夾中所有PDF文件的列表
pdf_files = [f for f in os.listdir(path) if f.endswith('.pdf')]
print(pdf_files)

# 遍歷每個PDF文件并使用Camelot讀取
for pdf_file in pdf_files:
    cross_page_tables = []
    print(f'開始對{pdf_file}進行表格識別')
    pdf_path = os.path.join(path, pdf_file)
    tables = camelot.read_pdf(pdf_path, pages='all')
    # 導(dǎo)出pdf所有的表格為json文件
    tables.export(f'{pdf_file}.json', f='json')  # json, excel, html, sqlite, markdown
    with pdfplumber.open(pdf_path) as pdf:
        for i, page in enumerate(pdf.pages[:-1]):
            table_ending, ending_col_count = is_table_ending_on_page(pdf_path, page, i)
            if i+ 1 < len(pdf.pages):
                table_starting, starting_col_count = is_table_starting_on_page(pdf_path, page, i + 1)
                if table_ending and table_starting and ending_col_count == starting_col_count:
                    cross_page_tables.append((i, i + 1))
                    print(f" {pdf_file}的表格在第 {i + 1} 頁和第 {i + 2} 頁之間跨頁，并且最后一行和下一頁的第一行列數(shù)相同")

        # 處理所有跨頁表格
        if cross_page_tables:
            current_merge = []
            for (start_page, end_page) in cross_page_tables:
                if not current_merge or start_page == current_merge[-1]:
                    current_merge.append(end_page)
                else:
                    # 合并當(dāng)前跨頁表格
                    merged_table = camelot.read_pdf(pdf_path, pages=",".join(map(str, [p + 1 for p in current_merge])))
                    full_df = pd.concat([table.df for table in merged_table])
                    full_df.to_csv(f'{pdf_file}_merged_{"_".join(map(str, [p + 1 for p in current_merge]))}.csv',
                                   index=False)
                    full_df.to_json(f'{pdf_file}_merged_{"_".join(map(str, [p + 1 for p in current_merge]))}.json',
                                    orient='records')
                    print(f'合并表格已保存為 {pdf_file}_merged_{"_".join(map(str, [p + 1 for p in current_merge]))}.csv 和 {pdf_file}_merged_{"_".join(map(str, [p + 1 for p in current_merge]))}.json')
                    # 重置合并列表
                    current_merge = [start_page, end_page]