pdf解析问题

[LimKim]

def parse_pdf(pdf_path: str, extract_image: bool = False) -> List[dict]:

    from pdfminer.high_level import extract_pages
    from pdfminer.layout import LTImage, LTRect, LTTextContainer
    doc = []
    import pdfplumber
    pdf = pdfplumber.open(pdf_path)
    for i, page_layout in enumerate(extract_pages(pdf_path)):
        page = {'page_num': page_layout.pageid, 'content': []}

        elements = []
        for element in page_layout:
            elements.append(element)

        # Init params for table
        table_num = 0
        tables = []

        for element in elements:
            if isinstance(element, LTRect):
                if not tables:
                    tables = extract_tables(pdf, i)
                if table_num < len(tables):
                    table_string = table_converter(tables[table_num])
                    table_num += 1
                    if table_string:
                        page['content'].append({'table': table_string, 'obj': element})
            elif isinstance(element, LTTextContainer):
                # Delete line breaks in the same paragraph
                text = element.get_text()
                # Todo: Further analysis using font
                font = get_font(element)
                if text.strip():
                    new_content_item = {'text': text, 'obj': element}
                    if font:
                        new_content_item['font-size'] = round(font[1])
                        # new_content_item['font-name'] = font[0]
                    page['content'].append(new_content_item)
            elif extract_image and isinstance(element, LTImage):
                # Todo: ocr
                raise ValueError('Currently, extracting images is not supported!')
            else:
                pass

        # merge elements
        page['content'] = postprocess_page_content(page['content'])
        doc.append(page)

    return doc

对于专利pdf，pdfminer 的 layout 分析器（LTTextContainer）无法正确聚合该 PDF 中的文本字符，导致正文被拆成大量孤立、无语义的碎片（如单个字母、数字、空格），使得后续的合并、去重、结构识别全部失效