优化OCR解析

rag/flow/splitter/splitter.py

@@ -19,7 +19,7 @@ import trio
 
 from api.utils import get_uuid
 from api.utils.base64_image import id2image, image2id
-from deepdoc.parser.pdf_parser import RAGFlowPdfParser
+from ocr.service import get_ocr_service
 from rag.flow.base import ProcessBase, ProcessParamBase
 from rag.flow.splitter.schema import SplitterFromUpstream
 from rag.nlp import naive_merge, naive_merge_with_images
@@ -96,14 +96,18 @@ class Splitter(ProcessBase):
             deli,
             self._param.overlapped_percent,
         )
-        cks = [
-            {
-                "text": RAGFlowPdfParser.remove_tag(c),
+        ocr_service = get_ocr_service()
+        cks = []
+        for c, img in zip(chunks, images):
+            if not c.strip():
+                continue
+            cleaned_text = await ocr_service.remove_tag(c)
+            positions = await ocr_service.extract_positions(c)
+            cks.append({
+                "text": cleaned_text,
                 "image": img,
-                "positions": [[pos[0][-1]+1, *pos[1:]] for pos in RAGFlowPdfParser.extract_positions(c)],
-            }
-            for c, img in zip(chunks, images) if c.strip()
-        ]
+                "positions": [[pos[0][-1]+1, *pos[1:]] for pos in positions],
+            })
         async with trio.open_nursery() as nursery:
             for d in cks:
                 nursery.start_soon(image2id, d, partial(STORAGE_IMPL.put), get_uuid())