feat: OCR 服务 + 会议材料模块
ry-ocr/ (新)
本地发票识别微服务 (PaddleOCR 3.x + FastAPI, 8801)
- QR 优先: 扫到二维码即取开票时间/发票号/金额; 没扫到/格式不合法直接判非发票, 不跑 OCR
- 配置 QR_FULL_OCR 控制快路径(false, 0.2s)还是全字段(true, 4.5s)
- /recognize/invoice (multipart) + /recognize/invoice/by-path (本地路径, 白名单) + /recognize/text
- is_invoice / from_qr / qr_raw / qr_error / error_code 字段
- 12 字段发票抽取 (regex + 启发式, 左右主体识别)
- 超时保护 (15s 单页 / 60s 总流程) + PaddleOCR 单例 + ThreadPoolExecutor
ry-api/ruoyi-business/
- pom.xml: 加 hutool-http/json/core 5.8.27, lombok 1.18.30 (OcrClient @Slf4j 所需)
- ocr/: OcrClient + InvoiceResult/Fields/Line + ZipExtractor + InvoiceOcrScheduler
- oss/: OssUploader + OssConfMeta (OCR 识别后重传 OSS)
- config/: OcrConfig + OcrExecutorConfig (后台线程池)
- service/impl/InvoiceOcrService: 后台提交 OCR, ZIP 路径解压识别, 替换场景先清旧
- 会议材料 CRUD 全套 (BizMeetingAuditLog/Executor/Invoice/Material/Supervisor):
controller + service + mapper + domain + xml
ry-vue3/
- MeetingDetail.vue (新建): 会议详情页 (含评分维度章节, 改只读)
- Meetings.vue / OssFileUploader.vue / router / Login.vue: 适配新字段
ry-api/ruoyi-admin/
- RuoYiApplication.java + application.yml: 启用 @Async 异步支持
_self/
- manager_meetings.md / manager_meeting_detail.md: 文档
This commit is contained in:
郭庆泰
@@ -0,0 +1,10 @@
|
||||
"""业务服务"""
|
||||
from .invoice_extractor import extract as extract_invoice
|
||||
from .qr_decoder import decode_qr, QRDecodeResult, has_any_field as qr_has_any_field
|
||||
|
||||
__all__ = [
|
||||
"extract_invoice",
|
||||
"decode_qr",
|
||||
"QRDecodeResult",
|
||||
"qr_has_any_field",
|
||||
]
|
||||
@@ -0,0 +1,299 @@
|
||||
"""从 OCR 文本/行里抽取发票字段
|
||||
|
||||
适配中国大陆 增值税发票（电子普票 / 专票 / 电子专票 / 数电票）
|
||||
|
||||
关键策略：
|
||||
- 主体（销售方/购买方）按 OCR box **坐标**判断归属（左右两栏）
|
||||
- 名称提取加 stop word，避免 OCR 单行文本混淆
|
||||
"""
|
||||
from __future__ import annotations
|
||||
|
||||
import re
|
||||
from typing import List, Optional, Tuple
|
||||
|
||||
from app.models import InvoiceFields, OCRLine
|
||||
from app.utils import (
|
||||
amount_consistent,
|
||||
extract_cn_amount,
|
||||
extract_num_amount,
|
||||
extract_pretax_amount,
|
||||
extract_tax_amount,
|
||||
extract_total_amount,
|
||||
)
|
||||
|
||||
|
||||
# ---------- 发票类型 ----------
|
||||
_INVOICE_TYPES = [
|
||||
"增值税电子专用发票",
|
||||
"增值税电子普通发票",
|
||||
"增值税专用发票",
|
||||
"增值税普通发票",
|
||||
"通用机打发票",
|
||||
"数电票",
|
||||
"电子发票",
|
||||
]
|
||||
|
||||
# ---------- 发票号码 ----------
|
||||
_NO_PATTERN = re.compile(
|
||||
r"(?:发\s*票\s*号\s*码|号\s*码|No\.?|号)\s*[:：]?\s*(\d{8,20})",
|
||||
re.IGNORECASE,
|
||||
)
|
||||
|
||||
# ---------- 发票代码 ----------
|
||||
_CODE_PATTERN = re.compile(
|
||||
r"(?:发\s*票\s*代\s*码|代\s*码)\s*[:：]?\s*(\d{10,12}|\d{8,12})",
|
||||
)
|
||||
|
||||
# ---------- 开票日期 ----------
|
||||
_DATE_PATTERN = re.compile(
|
||||
r"(?:开\s*票\s*日\s*期|日\s*期)\s*[:：]?\s*"
|
||||
r"(\d{4})\s*[年/\.]\s*(\d{1,2})\s*[月/\.]\s*(\d{1,2})",
|
||||
)
|
||||
|
||||
# ---------- 纳税人识别号（必须至少含 1 个字母，排除纯数字发票号）----------
|
||||
_TAX_NO_PATTERN = re.compile(r"((?=[0-9A-Z]*[A-Z])[0-9A-Z]{18})")
|
||||
|
||||
# ---------- 主体标签 ----------
|
||||
_BUYER_LABEL = re.compile(r"购\s*买\s*方\s*(?:信\s*息|名\s*称|)")
|
||||
_SELLER_LABEL = re.compile(r"销\s*售\s*方\s*(?:信\s*息|名\s*称|)")
|
||||
|
||||
# ---------- 名称（带 stop word 截断）----------
|
||||
_NAME_STOP = r"(?:销售方|购买方|统一社会信用|纳税人|项目名称|规格型号|^单位$|^数量$|^单价$|^金额|^税率|^税额|备注|收款人|复核|开票人|价税合计|小写|大写)"
|
||||
_NAME_PATTERN = re.compile(
|
||||
rf"名\s*称\s*[:：]\s*"
|
||||
rf"((?:(?!{_NAME_STOP})[^\n\r]){{2,60}}?(?:公司|商店|厂|店|部|中心|工作室))"
|
||||
)
|
||||
|
||||
|
||||
# ---------- 工具函数 ----------
|
||||
|
||||
def _norm(text: str) -> str:
|
||||
return re.sub(r"\s+", " ", text or "").strip()
|
||||
|
||||
|
||||
def _box_center(box) -> Tuple[float, float]:
|
||||
"""box: [[x1,y1], [x2,y2], [x3,y3], [x4,y4]] → (cx, cy)"""
|
||||
if not box or len(box) < 4:
|
||||
return (0.0, 0.0)
|
||||
xs = [p[0] for p in box]
|
||||
ys = [p[1] for p in box]
|
||||
return ((min(xs) + max(xs)) / 2, (min(ys) + max(ys)) / 2)
|
||||
|
||||
|
||||
def _detect_invoice_type(text: str) -> Optional[str]:
|
||||
for t in _INVOICE_TYPES:
|
||||
if t in text:
|
||||
return t
|
||||
return None
|
||||
|
||||
|
||||
def _extract_invoice_no(text: str) -> Optional[str]:
|
||||
m = _NO_PATTERN.search(text)
|
||||
return m.group(1) if m else None
|
||||
|
||||
|
||||
def _extract_invoice_code(text: str) -> Optional[str]:
|
||||
m = _CODE_PATTERN.search(text)
|
||||
return m.group(1) if m else None
|
||||
|
||||
|
||||
def _extract_date(text: str) -> Optional[str]:
|
||||
m = _DATE_PATTERN.search(text)
|
||||
if not m:
|
||||
return None
|
||||
y, mo, d = m.groups()
|
||||
return f"{int(y):04d}-{int(mo):02d}-{int(d):02d}"
|
||||
|
||||
|
||||
def _clean_name(name: str) -> str:
|
||||
name = re.sub(r"^[\s:：,，。、]+", "", name)
|
||||
name = name.split("纳税人")[0].split("统一社会")[0]
|
||||
name = re.sub(r"[^一-龥A-Za-z0-9（）()·\-]", "", name)
|
||||
return name.rstrip("：:;,，。、 ").strip()
|
||||
|
||||
|
||||
def _extract_parties_from_lines(lines: List[OCRLine]) -> tuple:
|
||||
"""基于 box 坐标的主体识别
|
||||
|
||||
1) 找到 "购买方信息"/"销售方信息" 标签所在位置（box 中心）
|
||||
2) 把每个 "名称: xxx" 行按 x 坐标归属到对应栏
|
||||
3) 兜底：左右栏无法区分时按 y 坐标（top/bottom）
|
||||
"""
|
||||
# 找标签坐标
|
||||
buyer_label_box = None
|
||||
seller_label_box = None
|
||||
for line in lines:
|
||||
if _BUYER_LABEL.search(line.text) and not buyer_label_box:
|
||||
buyer_label_box = line.box
|
||||
if _SELLER_LABEL.search(line.text) and not seller_label_box:
|
||||
seller_label_box = line.box
|
||||
|
||||
by_x = buyer_label_box is not None and seller_label_box is not None and \
|
||||
abs(_box_center(buyer_label_box)[0] - _box_center(seller_label_box)[0]) > 50
|
||||
|
||||
# 找所有 "名称: ..." 行的 box
|
||||
name_lines = []
|
||||
for line in lines:
|
||||
m = _NAME_PATTERN.search(line.text)
|
||||
if m:
|
||||
name_lines.append((line, _clean_name(m.group(1))))
|
||||
|
||||
# 找所有税号
|
||||
tax_lines = []
|
||||
for line in lines:
|
||||
m = _TAX_NO_PATTERN.search(line.text)
|
||||
if m:
|
||||
tax_lines.append((line, m.group(1)))
|
||||
|
||||
seller_name = buyer_name = None
|
||||
seller_tax = buyer_tax = None
|
||||
|
||||
if by_x and buyer_label_box and seller_label_box:
|
||||
# 左右栏布局：用 x 坐标归属
|
||||
bx, _ = _box_center(buyer_label_box)
|
||||
sx, _ = _box_center(seller_label_box)
|
||||
mid = (bx + sx) / 2
|
||||
|
||||
for line, name in name_lines:
|
||||
cx, _ = _box_center(line.box)
|
||||
if cx < mid and not buyer_name:
|
||||
buyer_name = name
|
||||
elif cx >= mid and not seller_name:
|
||||
seller_name = name
|
||||
|
||||
for line, tax in tax_lines:
|
||||
cx, _ = _box_center(line.box)
|
||||
if cx < mid and not buyer_tax:
|
||||
buyer_tax = tax
|
||||
elif cx >= mid and not seller_tax:
|
||||
seller_tax = tax
|
||||
else:
|
||||
# 上下栏 / 单栏：按 y 坐标（第一个 = 购买方）
|
||||
name_lines.sort(key=lambda x: (_box_center(x[0].box)[1], _box_center(x[0].box)[0]))
|
||||
tax_lines.sort(key=lambda x: (_box_center(x[0].box)[1], _box_center(x[0].box)[0]))
|
||||
if len(name_lines) >= 1:
|
||||
buyer_name = name_lines[0][1]
|
||||
if len(name_lines) >= 2:
|
||||
seller_name = name_lines[1][1]
|
||||
if len(tax_lines) >= 1:
|
||||
buyer_tax = tax_lines[0][1]
|
||||
if len(tax_lines) >= 2:
|
||||
seller_tax = tax_lines[1][1]
|
||||
|
||||
return seller_name, seller_tax, buyer_name, buyer_tax
|
||||
|
||||
|
||||
def _extract_parties_from_text(text: str) -> tuple:
|
||||
"""无 box 信息时的 fallback：按显式标签分段"""
|
||||
seller_name = buyer_name = None
|
||||
seller_tax = buyer_tax = None
|
||||
|
||||
for m in _BUYER_LABEL.finditer(text):
|
||||
chunk = text[m.end():m.end() + 200]
|
||||
nm = _NAME_PATTERN.search(chunk)
|
||||
tm = _TAX_NO_PATTERN.search(chunk[:100])
|
||||
if nm and not buyer_name:
|
||||
buyer_name = _clean_name(nm.group(1))
|
||||
if tm and not buyer_tax:
|
||||
buyer_tax = tm.group(1)
|
||||
if buyer_name and buyer_tax:
|
||||
break
|
||||
|
||||
for m in _SELLER_LABEL.finditer(text):
|
||||
chunk = text[m.end():m.end() + 200]
|
||||
nm = _NAME_PATTERN.search(chunk)
|
||||
tm = _TAX_NO_PATTERN.search(chunk[:100])
|
||||
if nm and not seller_name:
|
||||
seller_name = _clean_name(nm.group(1))
|
||||
if tm and not seller_tax:
|
||||
seller_tax = tm.group(1)
|
||||
if seller_name and seller_tax:
|
||||
break
|
||||
|
||||
if not buyer_name:
|
||||
name_positions = [(m.start(), m.group(1)) for m in _NAME_PATTERN.finditer(text)]
|
||||
if len(name_positions) >= 1:
|
||||
buyer_name = _clean_name(name_positions[0][1])
|
||||
if len(name_positions) >= 2:
|
||||
seller_name = _clean_name(name_positions[1][1])
|
||||
|
||||
tax_positions = _TAX_NO_PATTERN.findall(text)
|
||||
if not buyer_tax and len(tax_positions) >= 1:
|
||||
buyer_tax = tax_positions[0]
|
||||
if not seller_tax and len(tax_positions) >= 2:
|
||||
seller_tax = tax_positions[1]
|
||||
|
||||
return seller_name, seller_tax, buyer_name, buyer_tax
|
||||
|
||||
|
||||
def extract(text: str, lines: Optional[List[OCRLine]] = None) -> InvoiceFields:
|
||||
raw = _norm(text)
|
||||
|
||||
fields = InvoiceFields()
|
||||
fields.invoice_type = _detect_invoice_type(raw)
|
||||
fields.invoice_code = _extract_invoice_code(raw)
|
||||
fields.invoice_no = _extract_invoice_no(raw)
|
||||
fields.invoice_date = _extract_date(raw)
|
||||
|
||||
# 金额
|
||||
total = extract_total_amount(raw)
|
||||
tax = extract_tax_amount(raw)
|
||||
pretax = extract_pretax_amount(raw)
|
||||
|
||||
# 兜底：从所有 .xx 数字中找满足 tax + pretax = total 的组合
|
||||
if (tax is None or pretax is None) and total is not None:
|
||||
candidates = sorted(
|
||||
{float(x) for x in re.findall(r"(\d+\.\d{2})", raw) if float(x) < total},
|
||||
reverse=True,
|
||||
)
|
||||
for i, a in enumerate(candidates):
|
||||
for b in candidates[i + 1:]:
|
||||
if abs(a + b - total) < 0.011:
|
||||
if pretax is None:
|
||||
pretax = round(a, 2)
|
||||
if tax is None:
|
||||
tax = round(b, 2)
|
||||
break
|
||||
if tax is not None and pretax is not None:
|
||||
break
|
||||
# 单数字兜底：如果 total 不在候选里，只剩一个候选时
|
||||
if (tax is None or pretax is None) and len(candidates) == 1:
|
||||
only = round(candidates[0], 2)
|
||||
if pretax is None and tax is None:
|
||||
pretax = only
|
||||
tax = round(total - only, 2)
|
||||
elif tax is None:
|
||||
tax = only
|
||||
elif pretax is None:
|
||||
pretax = only
|
||||
|
||||
# 二次兜底：合计 - 任一 = 另一
|
||||
if tax is None and total is not None and pretax is not None:
|
||||
tax = round(total - pretax, 2)
|
||||
if pretax is None and total is not None and tax is not None:
|
||||
pretax = round(total - tax, 2)
|
||||
|
||||
fields.amount = total
|
||||
fields.tax_amount = tax
|
||||
fields.amount_pretax = pretax
|
||||
fields.amount_cn = extract_cn_amount(raw)
|
||||
fields.amount_match = amount_consistent(fields.amount_cn, fields.amount)
|
||||
|
||||
# 主体
|
||||
if lines:
|
||||
seller_name, seller_tax, buyer_name, buyer_tax = _extract_parties_from_lines(lines)
|
||||
if not (seller_name and buyer_name):
|
||||
sn, st, bn, bt = _extract_parties_from_text(raw)
|
||||
seller_name = seller_name or sn
|
||||
buyer_name = buyer_name or bn
|
||||
seller_tax = seller_tax or st
|
||||
buyer_tax = buyer_tax or bt
|
||||
else:
|
||||
seller_name, seller_tax, buyer_name, buyer_tax = _extract_parties_from_text(raw)
|
||||
|
||||
fields.seller_name = seller_name
|
||||
fields.seller_tax_no = seller_tax
|
||||
fields.buyer_name = buyer_name
|
||||
fields.buyer_tax_no = buyer_tax
|
||||
|
||||
return fields
|
||||
@@ -0,0 +1,138 @@
|
||||
"""电子发票二维码识别
|
||||
|
||||
国家税务总局规范的电子发票二维码内容格式 (8 字段逗号分隔):
|
||||
01,<type>,<invoice_code>,<invoice_no>,<amount>,<date>,<check_code>,<reserved>
|
||||
|
||||
例: 01,31,,24922000000006110014,39500.00,20240202,,A371
|
||||
"""
|
||||
from __future__ import annotations
|
||||
|
||||
from dataclasses import dataclass
|
||||
from pathlib import Path
|
||||
from typing import Optional, Tuple
|
||||
|
||||
import cv2
|
||||
import numpy as np
|
||||
from loguru import logger
|
||||
|
||||
|
||||
@dataclass
|
||||
class QRDecodeResult:
|
||||
"""二维码识别结果"""
|
||||
invoice_no: Optional[str] = None # 发票号码
|
||||
amount: Optional[float] = None # 金额（小写）
|
||||
invoice_date: Optional[str] = None # 开票日期 YYYY-MM-DD
|
||||
raw: str = "" # 二维码原始文本
|
||||
|
||||
|
||||
def _try_detect(img: np.ndarray, qd: cv2.QRCodeDetector) -> str:
|
||||
"""opencv 单图识别 + 多二维码识别, 任一成功即返回文本"""
|
||||
data, _, _ = qd.detectAndDecode(img)
|
||||
if data:
|
||||
return data
|
||||
try:
|
||||
retval, decoded_info, _, _ = qd.detectAndDecodeMulti(img)
|
||||
if decoded_info:
|
||||
# 取第一个非空的
|
||||
for d in decoded_info:
|
||||
if d:
|
||||
return d
|
||||
except Exception:
|
||||
pass
|
||||
return ""
|
||||
|
||||
|
||||
def _detect_qr(image_path: Path) -> str:
|
||||
"""从图片里解二维码; 全图 + 四象限各扫一遍"""
|
||||
img = cv2.imread(str(image_path))
|
||||
if img is None:
|
||||
return ""
|
||||
|
||||
qd = cv2.QRCodeDetector()
|
||||
# 1) 全图
|
||||
txt = _try_detect(img, qd)
|
||||
if txt:
|
||||
return txt
|
||||
|
||||
# 2) 四象限 (二维码常在票面边角)
|
||||
h, w = img.shape[:2]
|
||||
for name, crop in [
|
||||
("left-top", img[: h // 2, : w // 2]),
|
||||
("right-top", img[: h // 2, w // 2 :]),
|
||||
("left-bottom", img[h // 2 :, : w // 2]),
|
||||
("right-bottom", img[h // 2 :, w // 2 :]),
|
||||
]:
|
||||
txt = _try_detect(crop, qd)
|
||||
if txt:
|
||||
logger.debug("QR found in {}", name)
|
||||
return txt
|
||||
|
||||
# 3) 放大再试 (二维码像素过小的情况)
|
||||
scaled = cv2.resize(img, None, fx=2.0, fy=2.0, interpolation=cv2.INTER_CUBIC)
|
||||
return _try_detect(scaled, qd)
|
||||
|
||||
|
||||
def _parse_qr_payload(raw: str) -> QRDecodeResult:
|
||||
"""解析电子发票二维码内容 → QRDecodeResult
|
||||
|
||||
格式不合规返回空对象 (字段全 None), 由调用方判定为「格式不正确」并 fallback。
|
||||
"""
|
||||
if not raw:
|
||||
return QRDecodeResult(raw=raw or "")
|
||||
|
||||
parts = raw.split(",")
|
||||
if len(parts) != 8:
|
||||
logger.debug("QR 字段数 {} != 8, 视为格式不正确", len(parts))
|
||||
return QRDecodeResult(raw=raw)
|
||||
|
||||
result = QRDecodeResult(raw=raw)
|
||||
|
||||
# parts[3] = 发票号
|
||||
invoice_no = parts[3].strip()
|
||||
if invoice_no and (10 <= len(invoice_no) <= 30):
|
||||
result.invoice_no = invoice_no
|
||||
|
||||
# parts[4] = 金额
|
||||
amt_str = parts[4].strip()
|
||||
if amt_str:
|
||||
try:
|
||||
result.amount = round(float(amt_str), 2)
|
||||
except ValueError:
|
||||
pass
|
||||
|
||||
# parts[5] = 开票日期 (YYYYMMDD)
|
||||
date_str = parts[5].strip()
|
||||
if len(date_str) == 8 and date_str.isdigit():
|
||||
result.invoice_date = f"{date_str[:4]}-{date_str[4:6]}-{date_str[6:8]}"
|
||||
|
||||
return result
|
||||
|
||||
|
||||
def decode_qr(image_path: Path) -> QRDecodeResult:
|
||||
"""从图片文件解电子发票二维码
|
||||
|
||||
返回 QRDecodeResult; 字段全 None 表示无二维码或格式不正确。
|
||||
调用方应据此判定 fallback。
|
||||
"""
|
||||
try:
|
||||
raw = _detect_qr(image_path)
|
||||
except Exception as e:
|
||||
logger.warning("QR 检测异常 {}: {}", image_path.name, e)
|
||||
return QRDecodeResult()
|
||||
|
||||
if not raw:
|
||||
return QRDecodeResult()
|
||||
|
||||
parsed = _parse_qr_payload(raw)
|
||||
# 任一关键字段解出即视为「格式正确」
|
||||
if parsed.invoice_no or parsed.amount or parsed.invoice_date:
|
||||
logger.info("QR 解码成功: no={}, amt={}, date={}",
|
||||
parsed.invoice_no, parsed.amount, parsed.invoice_date)
|
||||
else:
|
||||
logger.debug("QR 解出但字段无效: raw={}", raw[:80])
|
||||
return parsed
|
||||
|
||||
|
||||
def has_any_field(qr: QRDecodeResult) -> bool:
|
||||
"""QR 是否解出了至少一个核心字段"""
|
||||
return bool(qr.invoice_no or qr.amount or qr.invoice_date)
|
||||
@@ -0,0 +1,237 @@
|
||||
"""端到端识别流水：文件 → QR → (可选)OCR → 字段抽取 → InvoiceResult
|
||||
|
||||
整体有 OCR_TOTAL_TIMEOUT_S 兜底；单页有 OCR_PAGE_TIMEOUT_S 兜底。
|
||||
任一超时立即返回失败，不再死等。
|
||||
|
||||
QR 流程 (config: QR_FULL_OCR):
|
||||
- 先扫二维码取 开票时间/发票号/金额 (国家税务总局 8 字段逗号规范)
|
||||
- 扫到 + qr_full_ocr=false → 直接返回这 3 个字段 (快路径, 跳过 OCR)
|
||||
- 扫到 + qr_full_ocr=true → 继续跑 OCR, QR 的 3 字段覆盖 OCR 抽取结果
|
||||
- 没扫到 / 格式不合法 → 直接判定为非发票 (不再跑 OCR)
|
||||
"""
|
||||
from __future__ import annotations
|
||||
|
||||
import shutil
|
||||
import tempfile
|
||||
import time
|
||||
from pathlib import Path
|
||||
from typing import List
|
||||
|
||||
from loguru import logger
|
||||
|
||||
from app.config import settings
|
||||
from app.core import OCRTimeout, auto_rotate, enhance, pdf_to_images, recognize
|
||||
from app.models import InvoiceResult, InvoiceFields, OCRLine
|
||||
from app.services import decode_qr, extract_invoice, qr_has_any_field
|
||||
|
||||
|
||||
_IMG_EXTS = {".png", ".jpg", ".jpeg", ".bmp", ".webp", ".tif", ".tiff"}
|
||||
|
||||
|
||||
def _save_upload(upload_bytes: bytes, suffix: str) -> Path:
|
||||
tmp = Path(tempfile.mkstemp(suffix=suffix)[1])
|
||||
tmp.write_bytes(upload_bytes)
|
||||
return tmp
|
||||
|
||||
|
||||
def _ocr_image(image_path: Path) -> List[OCRLine]:
|
||||
img = auto_rotate(image_path)
|
||||
img = enhance(img)
|
||||
return recognize(img)
|
||||
|
||||
|
||||
def _overlay_qr_fields(fields: InvoiceFields, qr) -> InvoiceFields:
|
||||
"""QR 解出的 3 字段优先, 没解到的保持 OCR 结果"""
|
||||
data = fields.model_dump()
|
||||
if qr.invoice_no:
|
||||
data["invoice_no"] = qr.invoice_no
|
||||
if qr.amount is not None:
|
||||
data["amount"] = qr.amount
|
||||
if qr.invoice_date:
|
||||
data["invoice_date"] = qr.invoice_date
|
||||
return InvoiceFields(**data)
|
||||
|
||||
|
||||
def _build_qr_only_result(qr, elapsed_ms: int, page_count: int) -> InvoiceResult:
|
||||
"""快路径: 只用 QR 字段, 跳过 OCR"""
|
||||
fields = InvoiceFields(
|
||||
invoice_no=qr.invoice_no,
|
||||
amount=qr.amount,
|
||||
invoice_date=qr.invoice_date,
|
||||
)
|
||||
return InvoiceResult(
|
||||
success=True,
|
||||
is_invoice=True,
|
||||
raw_text=f"[QR only] {qr.raw}",
|
||||
lines=[],
|
||||
fields=fields,
|
||||
page_count=page_count,
|
||||
engine="qr",
|
||||
elapsed_ms=elapsed_ms,
|
||||
from_qr=True,
|
||||
qr_raw=qr.raw,
|
||||
)
|
||||
|
||||
|
||||
def recognize_file(filename: str, content: bytes) -> InvoiceResult:
|
||||
"""识别入口（从字节流）：自动判断 PDF / 图片；任何环节超时立即返回失败"""
|
||||
t0 = time.time()
|
||||
suffix = Path(filename).suffix.lower()
|
||||
tmp_path = _save_upload(content, suffix or ".bin")
|
||||
try:
|
||||
return recognize_path(tmp_path, delete_after=True)
|
||||
except Exception:
|
||||
raise
|
||||
|
||||
|
||||
def recognize_path(file_path: Path, *, delete_after: bool = False) -> InvoiceResult:
|
||||
"""识别入口（从磁盘路径）：不做上传字节校验，直接读本地文件
|
||||
|
||||
Args:
|
||||
file_path: 已在白名单校验过的本地文件路径
|
||||
delete_after: True 表示临时文件用完删除（上传流场景），False 保留（原文件场景）
|
||||
"""
|
||||
t0 = time.time()
|
||||
suffix = file_path.suffix.lower()
|
||||
total_deadline = t0 + settings.ocr_total_timeout_s
|
||||
|
||||
def _elapsed_ms() -> int:
|
||||
return int((time.time() - t0) * 1000)
|
||||
|
||||
try:
|
||||
# ---------- 1. PDF / 图片 → 图片列表 ----------
|
||||
try:
|
||||
if suffix == ".pdf":
|
||||
page_imgs = pdf_to_images(file_path)
|
||||
elif suffix in _IMG_EXTS:
|
||||
page_imgs = [file_path]
|
||||
else:
|
||||
return InvoiceResult(
|
||||
success=False,
|
||||
is_invoice=False,
|
||||
error=f"不支持的文件类型: {suffix}（仅支持 PDF / 图片）",
|
||||
error_code="unsupported",
|
||||
engine="paddleocr",
|
||||
elapsed_ms=_elapsed_ms(),
|
||||
)
|
||||
except Exception as e:
|
||||
logger.exception("PDF/图片处理失败: {}", e)
|
||||
return InvoiceResult(
|
||||
success=False,
|
||||
is_invoice=False,
|
||||
error=f"PDF/图片处理失败: {e}",
|
||||
error_code="process_failed",
|
||||
engine="paddleocr",
|
||||
elapsed_ms=_elapsed_ms(),
|
||||
)
|
||||
|
||||
# ---------- 2. QR 优先识别 ----------
|
||||
qr = decode_qr(page_imgs[0])
|
||||
qr_ok = qr_has_any_field(qr)
|
||||
|
||||
if qr_ok and not settings.qr_full_ocr:
|
||||
# 快路径: 扫到 QR 且配置为 fast, 直接返回
|
||||
logger.info("QR 快路径: {} 耗时 {}ms", file_path.name, _elapsed_ms())
|
||||
return _build_qr_only_result(qr, _elapsed_ms(), len(page_imgs))
|
||||
|
||||
if not qr_ok:
|
||||
# 没扫到 / 格式不合法 → 直接认为非发票, 不再跑 OCR
|
||||
reason = "未识别到发票二维码" if not qr.raw else "二维码格式不合法"
|
||||
logger.info("非发票 (无有效 QR): {} 耗时 {}ms", file_path.name, _elapsed_ms())
|
||||
return InvoiceResult(
|
||||
success=False,
|
||||
is_invoice=False,
|
||||
error=f"{reason}（可能不是发票图片）",
|
||||
error_code="not_invoice",
|
||||
page_count=len(page_imgs),
|
||||
engine="paddleocr",
|
||||
elapsed_ms=_elapsed_ms(),
|
||||
from_qr=False,
|
||||
qr_raw=qr.raw or None,
|
||||
qr_error="no_qr" if not qr.raw else "bad_format",
|
||||
)
|
||||
|
||||
# ---------- 3. 每页 OCR（单页超时 + 总超时双重保护）----------
|
||||
# 到这里 qr_ok=True, OCR 仅为补充信息; QR 已保证是发票
|
||||
all_lines: List[OCRLine] = []
|
||||
for idx, img_path in enumerate(page_imgs, start=1):
|
||||
remaining = total_deadline - time.time()
|
||||
if remaining <= 0:
|
||||
logger.warning("达到总超时 ({:.1f}s), 中断 OCR", settings.ocr_total_timeout_s)
|
||||
return InvoiceResult(
|
||||
success=False,
|
||||
is_invoice=False,
|
||||
error=f"达到总超时 ({settings.ocr_total_timeout_s}秒), 已识别 {idx-1}/{len(page_imgs)} 页",
|
||||
error_code="timeout",
|
||||
raw_text="\n".join(l.text for l in all_lines),
|
||||
lines=all_lines,
|
||||
page_count=len(page_imgs),
|
||||
engine="paddleocr",
|
||||
elapsed_ms=_elapsed_ms(),
|
||||
from_qr=True,
|
||||
qr_raw=qr.raw,
|
||||
)
|
||||
page_timeout = min(settings.ocr_page_timeout_s, remaining)
|
||||
try:
|
||||
all_lines.extend(_ocr_image(img_path))
|
||||
except OCRTimeout as e:
|
||||
logger.warning("第 {} 页 OCR 超时: {}", idx, e)
|
||||
return InvoiceResult(
|
||||
success=False,
|
||||
is_invoice=False,
|
||||
error=f"第 {idx} 页识别超时 ({page_timeout:.1f}秒)",
|
||||
error_code="timeout",
|
||||
raw_text="\n".join(l.text for l in all_lines),
|
||||
lines=all_lines,
|
||||
page_count=len(page_imgs),
|
||||
engine="paddleocr",
|
||||
elapsed_ms=_elapsed_ms(),
|
||||
from_qr=True,
|
||||
qr_raw=qr.raw,
|
||||
)
|
||||
except Exception as e:
|
||||
logger.exception("第 {} 页 OCR 失败: {}", idx, e)
|
||||
return InvoiceResult(
|
||||
success=False,
|
||||
is_invoice=False,
|
||||
error=f"第 {idx} 页识别失败: {e}",
|
||||
error_code="ocr_failed",
|
||||
raw_text="\n".join(l.text for l in all_lines),
|
||||
lines=all_lines,
|
||||
page_count=len(page_imgs),
|
||||
engine="paddleocr",
|
||||
elapsed_ms=_elapsed_ms(),
|
||||
from_qr=True,
|
||||
qr_raw=qr.raw,
|
||||
)
|
||||
|
||||
# ---------- 4. 字段抽取 + QR 字段覆盖 ----------
|
||||
raw_text = "\n".join(l.text for l in all_lines)
|
||||
fields = extract_invoice(raw_text, all_lines)
|
||||
fields = _overlay_qr_fields(fields, qr)
|
||||
|
||||
elapsed = _elapsed_ms()
|
||||
logger.info("识别完成: {} 页={}, from_qr=true, 字段数={}, 耗时={}ms",
|
||||
file_path.name, len(page_imgs),
|
||||
sum(1 for f in fields.model_dump().values() if f), elapsed)
|
||||
return InvoiceResult(
|
||||
success=True,
|
||||
is_invoice=True,
|
||||
raw_text=raw_text,
|
||||
lines=all_lines,
|
||||
fields=fields,
|
||||
page_count=len(page_imgs),
|
||||
engine="paddleocr",
|
||||
elapsed_ms=elapsed,
|
||||
from_qr=True,
|
||||
qr_raw=qr.raw,
|
||||
)
|
||||
|
||||
finally:
|
||||
if delete_after:
|
||||
try:
|
||||
file_path.unlink(missing_ok=True)
|
||||
if suffix == ".pdf":
|
||||
shutil.rmtree(file_path.parent / f".{file_path.stem}_pages", ignore_errors=True)
|
||||
except Exception:
|
||||
pass
|
||||
Reference in New Issue
Block a user