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

ry-ocr/app/core/__init__.py

@@ -0,0 +1,6 @@
+"""核心能力：OCR / PDF / 图像"""
+from .ocr_engine import get_engine, recognize, warmup, OCRTimeout
+from .pdf_processor import pdf_to_images
+from .image_processor import auto_rotate, enhance
+
+__all__ = ["get_engine", "recognize", "warmup", "OCRTimeout", "pdf_to_images", "auto_rotate", "enhance"]

ry-ocr/app/core/image_processor.py

@@ -0,0 +1,47 @@
+"""图像预处理：自动旋转、放大、去噪"""
+from __future__ import annotations
+
+from pathlib import Path
+
+import cv2
+import numpy as np
+
+
+def auto_rotate(img_path: str | Path) -> Path:
+    """基于方向检测的简易旋转（> 阈值倾斜就转 90°）
+
+    说明：发票多为横向，此处只处理 0/90/180/270 四方向，
+    复杂倾斜交给 PaddleOCR 自带的 textline orientation。
+    """
+    p = Path(img_path)
+    img = cv2.imread(str(p))
+    if img is None:
+        return p
+
+    h, w = img.shape[:2]
+    # 横图宽 > 高 * 1.2，认为方向正确；否则旋转
+    if h > w * 1.2:
+        rotated = cv2.rotate(img, cv2.ROTATE_90_CLOCKWISE)
+        out = p.with_name(f"{p.stem}_rot.png")
+        cv2.imwrite(str(out), rotated)
+        return out
+    return p
+
+
+def enhance(img_path: str | Path) -> Path:
+    """轻度增强：灰度 + 自适应二值化（对手机拍的发票有帮助）"""
+    p = Path(img_path)
+    img = cv2.imread(str(p))
+    if img is None:
+        return p
+
+    gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
+    # 弱增强：只对低对比度图做二值化
+    if gray.std() < 50:
+        binary = cv2.adaptiveThreshold(
+            gray, 255, cv2.ADAPTIVE_THRESH_GAUSSIAN_C, cv2.THRESH_BINARY, 31, 10
+        )
+        out = p.with_name(f"{p.stem}_enh.png")
+        cv2.imwrite(str(out), binary)
+        return out
+    return p

ry-ocr/app/core/ocr_engine.py

@@ -0,0 +1,146 @@
+"""PaddleOCR 引擎封装（单例 + 超时）
+
+适配 paddleocr 3.0+ 的 predict() 接口。
+降级到 2.x 时也能跑（参数兼容）。
+"""
+from __future__ import annotations
+
+import threading
+from concurrent.futures import ThreadPoolExecutor, TimeoutError as FutTimeout
+from pathlib import Path
+from typing import List, Optional
+
+from loguru import logger
+
+from app.config import settings
+from app.models import OCRLine
+
+_lock = threading.Lock()
+_engine = None
+_executor = ThreadPoolExecutor(max_workers=2, thread_name_prefix="ocr")
+
+
+# ---------- 引擎构造 ----------
+
+def _build_engine():
+    try:
+        from paddleocr import PaddleOCR
+        # paddleocr >= 3.0: device="cpu"/"gpu", 没有 use_gpu / show_log
+        # 模型选择：
+        #   v5_server - 精度高但 CPU 慢 (5+ 秒/页)
+        #   v5_mobile - CPU 友好 (1~2 秒/页), 精度略低, 推荐生产环境
+        mobile = settings.ocr_lang.endswith("mobile") or settings.ocr_engine == "mobile"
+        if settings.use_gpu or settings.ocr_engine == "server":
+            # GPU 或用户显式指定 server 模型
+            det_name = "PP-OCRv5_server_det"
+            rec_name = "PP-OCRv5_server_rec"
+        else:
+            # 默认 mobile (CPU 友好)
+            det_name = "PP-OCRv5_mobile_det"
+            rec_name = "PP-OCRv5_mobile_rec"
+
+        return PaddleOCR(
+            use_doc_orientation_classify=False,
+            use_doc_unwarping=False,
+            use_textline_orientation=False,
+            lang=settings.ocr_lang,
+            device="gpu" if settings.use_gpu else "cpu",
+            text_detection_model_name=det_name,
+            text_recognition_model_name=rec_name,
+        )
+    except TypeError:
+        # paddleocr 2.x 老参数（向后兼容）
+        from paddleocr import PaddleOCR
+        return PaddleOCR(
+            use_angle_cls=True,
+            lang=settings.ocr_lang,
+            use_gpu=settings.use_gpu,
+            show_log=False,
+        )
+
+
+def get_engine():
+    global _engine
+    if _engine is None:
+        with _lock:
+            if _engine is None:
+                logger.info("正在初始化 PaddleOCR (lang={}, gpu={})...", settings.ocr_lang, settings.use_gpu)
+                _engine = _build_engine()
+                logger.info("PaddleOCR 初始化完成")
+    return _engine
+
+
+def _normalize(result) -> List[OCRLine]:
+    lines: List[OCRLine] = []
+    if not result:
+        return lines
+
+    if isinstance(result, list) and result and isinstance(result[0], dict):
+        for page in result:
+            texts = page.get("rec_texts") or []
+            scores = page.get("rec_scores") or []
+            polys = page.get("rec_polys") or []
+            for i, txt in enumerate(texts):
+                conf = float(scores[i]) if i < len(scores) else 0.0
+                box = polys[i].tolist() if i < len(polys) and hasattr(polys[i], "tolist") else []
+                lines.append(OCRLine(text=str(txt).strip(), confidence=conf, box=box))
+        return lines
+
+    try:
+        for page in result:
+            for det in page:
+                box = det[0]
+                txt, conf = det[1]
+                lines.append(OCRLine(text=str(txt).strip(), confidence=float(conf), box=box))
+    except Exception:
+        logger.exception("无法解析 PaddleOCR 输出: {}", result)
+    return lines
+
+
+class OCRTimeout(Exception):
+    """OCR 识别超时"""
+    pass
+
+
+def recognize(image_path: str | Path, timeout_s: Optional[float] = None) -> List[OCRLine]:
+    """对单张图片做 OCR，带超时控制
+
+    timeout_s 默认为 settings.ocr_page_timeout_s（默认 5 秒）。
+    超时立即抛 OCRTimeout，不等模型跑完。
+    """
+    timeout = timeout_s or settings.ocr_page_timeout_s
+    engine = get_engine()
+
+    future = _executor.submit(engine.predict, str(image_path))
+    try:
+        raw = future.result(timeout=timeout)
+    except FutTimeout:
+        future.cancel()
+        logger.error("OCR 超时 ({:.1f}s): {}", timeout, image_path)
+        raise OCRTimeout(f"OCR 识别超时 ({timeout}秒): {image_path}")
+    except Exception as e:
+        logger.exception("OCR 执行异常: {}", e)
+        raise
+
+    return _normalize(raw)
+
+
+def warmup():
+    """预热：构造引擎 + 真跑一次空白图预测（避免首次请求超时）
+
+    跳过预测失败（比如没装 paddle）但不报错。
+    """
+    try:
+        engine = get_engine()
+    except Exception as e:
+        logger.warning("OCR 引擎初始化失败: {}", e)
+        return
+
+    try:
+        import numpy as np
+        # 1x1 灰度图，足以触发模型完整链路
+        dummy = np.zeros((64, 64, 3), dtype=np.uint8)
+        engine.predict(dummy)
+        logger.info("OCR 预热完成（首次预测已跑）")
+    except Exception as e:
+        logger.warning("OCR 预测预热失败（不影响主流程）: {}", e)

ry-ocr/app/core/pdf_processor.py

@@ -0,0 +1,35 @@
+"""PDF → 图片 (使用 PyMuPDF，无需 poppler 系统依赖)"""
+from __future__ import annotations
+
+from pathlib import Path
+from typing import List
+
+import fitz  # PyMuPDF
+from loguru import logger
+
+from app.config import settings
+
+
+def pdf_to_images(pdf_path: str | Path, dpi: int | None = None) -> List[Path]:
+    """把 PDF 每页渲染成 PNG，返回文件路径列表"""
+    dpi = dpi or settings.pdf_dpi
+    pdf_path = Path(pdf_path)
+    out_dir = pdf_path.parent / f".{pdf_path.stem}_pages"
+    out_dir.mkdir(exist_ok=True)
+
+    doc = fitz.open(str(pdf_path))
+    saved: List[Path] = []
+    try:
+        # 1.0 = 72 DPI 的 zoom 系数
+        zoom = dpi / 72.0
+        mat = fitz.Matrix(zoom, zoom)
+        for idx, page in enumerate(doc, start=1):
+            pix = page.get_pixmap(matrix=mat, alpha=False)
+            out_path = out_dir / f"page_{idx:03d}.png"
+            pix.save(str(out_path))
+            saved.append(out_path)
+    finally:
+        doc.close()
+
+    logger.info("PDF 转图片: {} → {} 页 (dpi={})", pdf_path.name, len(saved), dpi)
+    return saved