优化OCR解析

rag/flow/hierarchical_merger/hierarchical_merger.py

@@ -22,7 +22,7 @@ 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.hierarchical_merger.schema import HierarchicalMergerFromUpstream
 from rag.nlp import concat_img
@@ -170,14 +170,17 @@ class HierarchicalMerger(ProcessBase):
                 cks.append(txt)
                 images.append(img)
 
-            cks = [
-                {
-                    "text": RAGFlowPdfParser.remove_tag(c),
+            ocr_service = get_ocr_service()
+            processed_cks = []
+            for c, img in zip(cks, images):
+                cleaned_text = await ocr_service.remove_tag(c)
+                positions = await ocr_service.extract_positions(c)
+                processed_cks.append({
+                    "text": cleaned_text,
                     "image": img,
-                    "positions": RAGFlowPdfParser.extract_positions(c),
-                }
-                for c, img in zip(cks, images)
-            ]
+                    "positions": positions,
+                })
+            cks = processed_cks
             async with trio.open_nursery() as nursery:
                 for d in cks:
                     nursery.start_soon(image2id, d, partial(STORAGE_IMPL.put), get_uuid())