a

personal-blog/backend/routers/universal_crawl.py

@@ -440,6 +440,8 @@ def _legacy_extract(url: str, html: str, images: Optional[list] = None):
         cto_html = _extract_51cto_content(html)
         markdown_content = html_to_markdown(cto_html, url)
     elif is_spa_site(html):
+        # 正文区 vs 噪音过滤（Pruning 思路）统一在 crawl.py 的
+        # convert_spa_html_to_markdown / html_to_markdown 入口处理，此处不重复剪枝
         markdown_content = convert_spa_html_to_markdown(html, images, url)
     else:
         markdown_content = html_to_markdown(html, url)
@@ -488,6 +490,13 @@ async def universal_fetch(request: UniversalCrawlRequest):
         else:
             title, markdown_content, images = _legacy_extract(url, html)
 
+        # 5.1 只保留正文中实际出现的图片（全页提取会带入头像/侧栏/推荐位等噪音图）
+        # 正文无图（如该掘金文章）时 images 会被清空，不再下载噪音图
+        def _norm_img_url(u: str) -> str:
+            return u.replace('&', '&')
+        markdown_images = {_norm_img_url(img.original_url) for img in extract_images_from_markdown(markdown_content, url)}
+        images = [img for img in images if _norm_img_url(img.original_url) in markdown_images]
+
         # 6. 数学公式修复
         markdown_content = fix_math_formulas(markdown_content)
 
@@ -563,13 +572,12 @@ async def universal_save(request: CrawlSaveRequest, db: Session = Depends(get_db
                     # 保存到数据库
                     db_image = Image(
                         filename=final_filename,
-                        original_filename=final_filename,
+                        original_name=final_filename,
                         file_path=f"/uploads/images/{final_filename}",
                         file_size=os.path.getsize(str(final_path)),
-                        mime_type=f"image/{os.path.splitext(final_filename)[1].lstrip('.') or 'jpeg'}",
                         width=width,
                         height=height,
-                        thumbnail_path=thumb_path_db,
+                        thumb_path=thumb_path_db,
                     )
                     db.add(db_image)
                     db.flush()