gx

core/detect_plate_process.py

@@ -16,7 +16,7 @@ class Config:
     # 你的Plate检测模型路径（相对项目根目录）；若存在 best_openvino_model 会自动优先加载
     MODEL_PATH = str(_ROOT / "best.pt")
 
-    CONF_THRESHOLD = 0.8
+    CONF_THRESHOLD = 0.90
     IOU_THRESHOLD = 0.45
     # 显式 imgsz，便于 OpenVINO / CPU 控制耗时；可用环境变量 DISH_YOLO_IMGSZ 覆盖
     IMGSZ = int(os.environ.get("DISH_YOLO_IMGSZ", "640"))

core/pending_learning_ui.py

@@ -31,13 +31,43 @@ def format_predict_detail(detail):
 
 
 def format_dish_box_label(detail):
-    """将识别结果格式化为图片框上的菜品名称标签。"""
+    """将识别结果格式化为图片框标签。
+
+    默认只显示菜名；设环境变量 DISH_SHOW_BOX_SCORES=1 时追加「似xx 检xx」。
+    """
     if not detail or detail.get("error"):
         return "识别失败"
+
     if detail.get("is_unknown"):
         suggested = detail.get("suggested_category")
-        return f"未知·{suggested}" if suggested else "未知"
-    return detail.get("label") or "未知"
+        base = f"未知·{suggested}" if suggested else "未知"
+    else:
+        base = detail.get("label") or detail.get("name") or "未知"
+
+    try:
+        from dao.settings import DISH_SHOW_BOX_SCORES
+        show_scores = bool(DISH_SHOW_BOX_SCORES)
+    except Exception:
+        show_scores = False
+    if not show_scores:
+        return str(base)
+
+    def _fmt(v):
+        try:
+            return f"{float(v):.2f}" if v is not None else None
+        except (TypeError, ValueError):
+            return None
+
+    sim = _fmt(detail.get("score"))
+    det = _fmt(detail.get("det_conf"))
+    bits = []
+    if sim is not None:
+        bits.append(f"似{sim}")
+    if det is not None:
+        bits.append(f"检{det}")
+    if not bits:
+        return str(base)
+    return f"{base} {' '.join(bits)}"
 
 
 class PendingLearningUIMixin:

dao/settings.py

@@ -71,6 +71,9 @@ DISH_DEVICE_STATUS_ENABLED = os.environ.get('DISH_DEVICE_STATUS_ENABLED', '1') =
 DISH_SYNC_ON_START = os.environ.get('DISH_SYNC_ON_START', '1') == '1'
 DISH_SYNC_INTERVAL = int(os.environ.get('DISH_SYNC_INTERVAL', 300))  # 默认 5 分钟
 
+# 识别框是否显示「似xx 检xx」（默认关；需要时 set DISH_SHOW_BOX_SCORES=1）
+DISH_SHOW_BOX_SCORES = os.environ.get('DISH_SHOW_BOX_SCORES', '0') == '1'
+
 DEBUG = True
 
 

ui/classifier_ui.py

@@ -1759,12 +1759,8 @@ class BaseDishApp(PendingLearningUIMixin):
             print(f"[live] preview failed: {e}")
 
     def _live_dish_label(self, dish):
-        if not dish:
-            return "未知"
-        if dish.get("is_unknown"):
-            sug = (dish.get("suggested_category") or "").strip()
-            return f"未知·{sug}" if sug else "未知"
-        return (dish.get("label") or dish.get("name") or "未知").strip() or "未知"
+        from ui.pending_learning_ui import format_dish_box_label
+        return format_dish_box_label(dish)
 
     def _draw_live_overlay_pil(self, pil_im, dishes, sx=1.0, sy=1.0):
         """在已缩放预览图上画识别框：整图一次 alpha 合成，避免逐框卡顿。"""
@@ -1967,6 +1963,7 @@ class BaseDishApp(PendingLearningUIMixin):
                 boxes,
                 auto_pending=False,
                 crop_images=crop_arrays,
+                det_confs=[c.get("conf") for c in crop_items],
             )
             self._last_detect_fp = fp
             self._last_live_dishes = list(dishes or [])
@@ -2404,6 +2401,7 @@ class BaseDishApp(PendingLearningUIMixin):
                     f"正在识别第 {cur} 张…", "busy"
                 ),
                 crop_images=crop_arrays,
+                det_confs=[c.get("conf") for c in crop_items],
             )
             preview = labeled_path or visualize_save_path
             if preview:
@@ -2523,7 +2521,9 @@ class BaseDishApp(PendingLearningUIMixin):
                     return
 
                 final_result, _, _unknown_count, labeled_path, dishes = self.run_plate_predict_batch(
-                    save_path, cropped_file_paths, boxes, crop_images=crop_arrays,
+                    save_path, cropped_file_paths, boxes,
+                    crop_images=crop_arrays,
+                    det_confs=[c.get("conf") for c in crop_items],
                 )
                 preview = labeled_path or visualize_save_path
                 self.report_recognition(save_path, dishes, preview)

ui/pending_learning_ui.py

@@ -31,13 +31,43 @@ def format_predict_detail(detail):
 
 
 def format_dish_box_label(detail):
-    """将识别结果格式化为图片框上的菜品名称标签。"""
+    """将识别结果格式化为图片框标签。
+
+    默认只显示菜名；设环境变量 DISH_SHOW_BOX_SCORES=1 时追加「似xx 检xx」。
+    """
     if not detail or detail.get("error"):
         return "识别失败"
+
     if detail.get("is_unknown"):
         suggested = detail.get("suggested_category")
-        return f"未知·{suggested}" if suggested else "未知"
-    return detail.get("label") or "未知"
+        base = f"未知·{suggested}" if suggested else "未知"
+    else:
+        base = detail.get("label") or detail.get("name") or "未知"
+
+    try:
+        from dao.settings import DISH_SHOW_BOX_SCORES
+        show_scores = bool(DISH_SHOW_BOX_SCORES)
+    except Exception:
+        show_scores = False
+    if not show_scores:
+        return str(base)
+
+    def _fmt(v):
+        try:
+            return f"{float(v):.2f}" if v is not None else None
+        except (TypeError, ValueError):
+            return None
+
+    sim = _fmt(detail.get("score"))
+    det = _fmt(detail.get("det_conf"))
+    bits = []
+    if sim is not None:
+        bits.append(f"似{sim}")
+    if det is not None:
+        bits.append(f"检{det}")
+    if not bits:
+        return str(base)
+    return f"{base} {' '.join(bits)}"
 
 
 class PendingLearningUIMixin:
@@ -689,6 +719,7 @@ class PendingLearningUIMixin:
     def run_plate_predict_batch(
         self, source_image_path, cropped_file_paths, boxes,
         progress_callback=None, auto_pending=True, crop_images=None,
+        det_confs=None,
     ):
         """对多个裁剪区域识别。
 
@@ -698,6 +729,7 @@ class PendingLearningUIMixin:
 
         cropped_file_paths: 裁剪图路径列表（可与 crop_images 二选一/并用）
         crop_images: 可选 BGR ndarray 列表，优先用于特征提取（免读盘）
+        det_confs: 可选 YOLO 检测置信度列表（与 boxes 对齐）
 
         返回:
             final_result, dish_labels, unknown_count, labeled_path, dishes
@@ -728,6 +760,7 @@ class PendingLearningUIMixin:
             if progress_callback:
                 progress_callback(i + 1, n)
             box = boxes[i] if boxes and i < len(boxes) else None
+            det_conf = det_confs[i] if det_confs and i < len(det_confs) else None
             img_path = cropped_file_paths[i] if i < len(cropped_file_paths) else None
             try:
                 detail = self.manager.predict_from_feature(
@@ -737,12 +770,15 @@ class PendingLearningUIMixin:
                 )
                 if detail.get("is_unknown"):
                     unknown_count += 1
-                dish_labels.append(format_dish_box_label(detail))
+                label_src = dict(detail)
+                label_src["det_conf"] = det_conf
+                dish_labels.append(format_dish_box_label(label_src))
                 final_result += f"【菜品{i + 1}】{format_predict_detail(detail)}\n"
                 dishes.append({
                     "index": i + 1,
                     "label": detail.get("label"),
                     "score": detail.get("score"),
+                    "det_conf": det_conf,
                     "is_unknown": bool(detail.get("is_unknown")),
                     "suggested_category": detail.get("suggested_category"),
                     "pending_id": detail.get("pending_id"),
@@ -759,6 +795,7 @@ class PendingLearningUIMixin:
                     "index": i + 1,
                     "label": None,
                     "score": None,
+                    "det_conf": det_conf,
                     "is_unknown": True,
                     "suggested_category": None,
                     "pending_id": None,