Feat: enhance Excel image extraction with vision-based descriptions (#12054)

### What problem does this PR solve?
issue:
[#11618](https://github.com/infiniflow/ragflow/issues/11618)
change:
enhance Excel image extraction with vision-based descriptions

### Type of change

- [x] New Feature (non-breaking change which adds functionality)

deepdoc/parser/excel_parser.py

@@ -18,6 +18,7 @@ from io import BytesIO
 
 import pandas as pd
 from openpyxl import Workbook, load_workbook
+from PIL import Image
 
 from rag.nlp import find_codec
 
@@ -109,6 +110,52 @@ class RAGFlowExcelParser:
                     ws.cell(row=row_num, column=col_num, value=value)
         return wb
 
+    @staticmethod
+    def _extract_images_from_worksheet(ws, sheetname=None):
+        """
+        Extract images from a worksheet and enrich them with vision-based descriptions.
+
+        Returns: List[dict]
+        """
+        images = getattr(ws, "_images", [])
+        if not images:
+            return []
+
+        raw_items = []
+
+        for img in images:
+            try:
+                img_bytes = img._data()
+                pil_img = Image.open(BytesIO(img_bytes)).convert("RGB")
+
+                anchor = img.anchor
+                if hasattr(anchor, "_from") and hasattr(anchor, "_to"):
+                    r1, c1 = anchor._from.row + 1, anchor._from.col + 1
+                    r2, c2 = anchor._to.row + 1, anchor._to.col + 1
+                    if r1 == r2 and c1 == c2:
+                        span = "single_cell"
+                    else:
+                        span = "multi_cell"
+                else:
+                    r1, c1 = anchor._from.row + 1, anchor._from.col + 1
+                    r2, c2 = r1, c1
+                    span = "single_cell"
+
+                item = {
+                    "sheet": sheetname or ws.title,
+                    "image": pil_img,
+                    "image_description": "",
+                    "row_from": r1,
+                    "col_from": c1,
+                    "row_to": r2,
+                    "col_to": c2,
+                    "span_type": span,
+                }
+                raw_items.append(item)
+            except Exception:
+                continue
+        return raw_items
+
     def html(self, fnm, chunk_rows=256):
         from html import escape
 

deepdoc/parser/figure_parser.py

@@ -55,6 +55,31 @@ def vision_figure_parser_docx_wrapper(sections, tbls, callback=None,**kwargs):
             callback(0.8, f"Visual model error: {e}. Skipping figure parsing enhancement.")
     return tbls
 
+def vision_figure_parser_figure_xlsx_wrapper(images,callback=None, **kwargs):
+    tbls = []
+    if not images:
+        return []
+    try:
+        vision_model = LLMBundle(kwargs["tenant_id"], LLMType.IMAGE2TEXT)
+        callback(0.2, "Visual model detected. Attempting to enhance Excel image extraction...")
+    except Exception:
+        vision_model = None
+    if vision_model:
+        figures_data = [((
+                        img["image"],   # Image.Image
+                        [img["image_description"]]     # description list (must be list)
+                    ),
+                    [
+                        (0, 0, 0, 0, 0)   # dummy position
+                    ]) for img in images]
+        try:
+            parser = VisionFigureParser(vision_model=vision_model, figures_data=figures_data, **kwargs)
+            callback(0.22, "Parsing images...")
+            boosted_figures = parser(callback=callback)
+            tbls.extend(boosted_figures)
+        except Exception as e:
+            callback(0.25, f"Excel visual model error: {e}. Skipping vision enhancement.")
+    return tbls
 
 def vision_figure_parser_pdf_wrapper(tbls, callback=None, **kwargs):
     if not tbls: