YG-Soft/X-Financial
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

feat(ocr): PDF 文本层可用时跳过 worker 调用并补装 poppler-data

Browse Source 
- OcrService 提取 PDF 文本层后若有效字符达到阈值，直接构建文档并写入结果缓存，不再触发 OCR worker，仅无文本层时才解析 python_bin/worker_path 调用 worker
- _build_text_layer_document 复用 AggregatedOcrDocument 聚合文本层片段，_has_usable_pdf_text_layer 基于 meaningful_char_count 判定
- docker-compose 与 paddleocr bootstrap 脚本补装 poppler-data，保证 PDF 文本层抽取的中文编码正确
- 新增文本层直取与运行时依赖两项 ocr_service 单测
This commit is contained in:
caoxiaozhu
2026-06-21 23:23:59 +08:00
parent 1986b0d945
commit 88e91a5900
5 changed files with 123 additions and 7 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
server/scripts/bootstrap_paddleocr_gpu.sh
View File

@@ -14,7 +14,7 @@ if ! command -v "${PYTHON_BIN}" >/dev/null 2>&1; then
fi
apt-get update
apt-get install -y --no-install-recommends libgl1 libglib2.0-0 poppler-utils
apt-get install -y --no-install-recommends libgl1 libglib2.0-0 poppler-utils poppler-data
rm -rf "${OCR_VENV_DIR}"
"${PYTHON_BIN}" -m venv "${OCR_VENV_DIR}"

2
server/scripts/bootstrap_paddleocr_mobile.sh
View File

@@ -13,7 +13,7 @@ if ! command -v "${PYTHON_BIN}" >/dev/null 2>&1; then
fi
apt-get update
apt-get install -y --no-install-recommends libgl1 libglib2.0-0 poppler-utils
apt-get install -y --no-install-recommends libgl1 libglib2.0-0 poppler-utils poppler-data
"${PYTHON_BIN}" -m venv "${OCR_VENV_DIR}"
"${OCR_VENV_DIR}/bin/pip" install --upgrade pip

55
server/src/app/services/ocr.py
View File

@@ -77,8 +77,6 @@ class OcrService:
documents: list[OcrRecognizeDocumentRead] = []
prepared_inputs: list[PreparedOcrInput] = []
cleanup_paths: list[Path] = []
python_bin = self._resolve_python_bin()
worker_path = self._resolve_worker_path()
worker_payload: dict = {}
cache_keys_by_source: dict[str, str] = {}
@@ -144,6 +142,16 @@ class OcrService:
cleanup_paths=cleanup_paths,
text_layer=text_layer,
)
if self._has_usable_pdf_text_layer(text_layer):
document = self._build_text_layer_document(
filename=normalized_name,
media_type=resolved_media_type,
text_layer=text_layer,
pdf_inputs=pdf_inputs,
)
documents.append(document)
self._write_cached_document(cache_key, document)
continue
prepared_inputs.extend(pdf_inputs)
for item in pdf_inputs:
cache_keys_by_source.setdefault(item.source_key, cache_key)
@@ -175,6 +183,8 @@ class OcrService:
cache_keys_by_source[source_key] = cache_key
if prepared_inputs:
python_bin = self._resolve_python_bin()
worker_path = self._resolve_worker_path()
worker_payload = self._invoke_worker(
python_bin=python_bin,
worker_path=worker_path,
@@ -308,6 +318,23 @@ class OcrService:
while len(cls._result_cache) > OCR_RESULT_CACHE_LIMIT:
cls._result_cache.popitem(last=False)
@classmethod
def _write_cached_document(cls, cache_key: str, document: OcrRecognizeDocumentRead) -> None:
if not cache_key:
return
with cls._cache_lock:
cls._result_cache[cache_key] = document.model_copy(
update={
"receipt_id": "",
"receipt_status": "",
"receipt_preview_url": "",
"receipt_source_url": "",
}
)
cls._result_cache.move_to_end(cache_key)
while len(cls._result_cache) > OCR_RESULT_CACHE_LIMIT:
cls._result_cache.popitem(last=False)
@classmethod
def _resolve_worker_semaphore(cls, limit: int) -> threading.Semaphore:
normalized_limit = max(1, int(limit or 1))
@@ -568,6 +595,30 @@ class OcrService:
return documents
def _build_text_layer_document(
self,
*,
filename: str,
media_type: str,
text_layer: str,
pdf_inputs: list[PreparedOcrInput],
) -> OcrRecognizeDocumentRead:
first_input = pdf_inputs[0] if pdf_inputs else None
aggregated = AggregatedOcrDocument(
filename=filename,
media_type=media_type,
source_key=first_input.source_key if first_input is not None else uuid4().hex,
page_count=max(1, len(pdf_inputs)),
preview_kind=str(first_input.preview_kind if first_input is not None else ""),
preview_data_url=str(first_input.preview_data_url if first_input is not None else ""),
)
aggregated.text_layer_fragments.append(text_layer)
return self._finalize_document(aggregated)
@classmethod
def _has_usable_pdf_text_layer(cls, text_layer: str) -> bool:
return cls._meaningful_char_count(text_layer) >= 8
@staticmethod
def _collect_descriptor_text_layer(descriptors: list[PreparedOcrInput]) -> str:
for descriptor in descriptors:

69
server/tests/test_ocr_service.py
View File

@@ -8,6 +8,18 @@ from app.core.config import get_settings
from app.services.ocr import OcrService
def test_ocr_runtime_installers_include_poppler_cjk_data() -> None:
repo_root = Path(__file__).resolve().parents[2]
dependency_sources = [
repo_root / "docker-compose.yml",
repo_root / "server" / "scripts" / "bootstrap_paddleocr_mobile.sh",
repo_root / "server" / "scripts" / "bootstrap_paddleocr_gpu.sh",
]
for path in dependency_sources:
assert "poppler-data" in path.read_text(encoding="utf-8")
def test_ocr_service_uses_worker_runtime_and_keeps_unsupported_files_as_warnings(
monkeypatch,
tmp_path: Path,
@@ -220,6 +232,59 @@ def test_ocr_service_converts_pdf_to_images_and_returns_image_preview(
assert recognized.lines[1].page_index == 1
def test_ocr_service_uses_pdf_text_layer_without_worker_runtime(
monkeypatch,
tmp_path: Path,
) -> None:
def fake_convert_pdf_to_images(self, *, pdf_path: Path, output_dir: Path) -> list[Path]:
page = output_dir / "page-1.png"
page.write_bytes(b"fake-rendered-page")
return [page]
def fail_resolve_python(self) -> str:
raise AssertionError("PDF 文本层可用时不应强制解析 OCR worker。")
def fail_invoke_worker(self, **kwargs) -> dict:
raise AssertionError("PDF 文本层可用时不应调用 OCR worker。")
monkeypatch.setenv("STORAGE_ROOT_DIR", str(tmp_path / "storage"))
monkeypatch.setattr(OcrService, "_resolve_python_bin", fail_resolve_python)
monkeypatch.setattr(OcrService, "_resolve_worker_path", lambda self: "worker.py")
monkeypatch.setattr(OcrService, "_convert_pdf_to_images", fake_convert_pdf_to_images)
monkeypatch.setattr(OcrService, "_invoke_worker", fail_invoke_worker)
monkeypatch.setattr(
OcrService,
"_extract_pdf_text_layer",
lambda self, pdf_path: (
"电子发票（铁路电子客票）\n"
"发票号码:26429165800002785705\n"
"武汉站\n"
"上海虹桥站\n"
"G458\n"
"票价:￥354.00\n"
"电子客票号:6580061086021391007342026"
),
)
get_settings.cache_clear()
try:
result = OcrService().recognize_files(
[
("train-ticket.pdf", b"%PDF-1.7 fake", "application/pdf"),
]
)
finally:
get_settings.cache_clear()
recognized = result.documents[0]
assert result.success_count == 1
assert recognized.document_type == "train_ticket"
assert "电子发票（铁路电子客票）" in recognized.text
assert "电子客票号:6580061086021391007342026" in recognized.text
assert any(field.label == "金额" and field.value == "354元" for field in recognized.document_fields)
assert recognized.preview_kind == "image"
assert recognized.preview_data_url.startswith("data:image/png;base64,")
def test_ocr_service_reuses_cached_document_for_same_content(
monkeypatch,
tmp_path: Path,
@@ -351,5 +416,5 @@ def test_ocr_service_prefers_pdf_text_layer_when_rendered_ocr_is_placeholder_hea
assert "上海虹桥站" in recognized.text
assert "□□□□" not in recognized.summary
assert recognized.document_type == "train_ticket"
assert recognized.preview_kind == ""
assert recognized.preview_data_url == ""
assert recognized.preview_kind == "image"
assert recognized.preview_data_url.startswith("data:image/png;base64,")